Columbia's Chan Tells Boards to Audit Near-Misses as AI Threats Mount

Panel 1 - 1: Tomorrow’s Cyber Risks Today — Ai4

Ai4

Perimeter defense cannot stop agentic AI and quantum threats, forcing organizations to build resilience by analyzing failed intrusions and mandating independent audit oversight directly to boards.

AI-assisted summary of Panel 1 - 1: Tomorrow’s Cyber Risks Today — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by the Ai4 channel on August 14, 2026, this presentation features Kitty Chan, Professor of Practice at Columbia University. According to the description, the session explores how artificial intelligence, agentic systems, and quantum technologies challenge conventional governance and oversight, presenting a framework for internal audit to evaluate and build organizational cyber resilience against evolving threats.

## TL;DR Chan argues that because modern threats like agentic AI exploits and quantum decryption make breaches inevitable, organizations must pivot from rigid perimeter prevention to holistic cyber resilience. Utilizing the NIST Cybersecurity Framework 2.0, enterprises must align operational business units, enterprise risk oversight, and independent internal audit across the three lines of defense. The presentation emphasizes proactive measures such as identifying 'Shadow AI,' systematically analyzing near misses and failed intrusions, and establishing fail-safe containment protocols with sequenced recovery plans. Ultimately, maintaining organizational resilience and regulatory confidence requires independent board-level assurance and a collaborative culture where security is embraced as a shared responsibility.

## Atmosphere & style The presentation carries a pragmatic, authoritative, and educational tone that replaces typical cybersecurity alarmism with structured, clear-eyed methodology. Chan speaks with the calm clarity of a seasoned professor and practitioner, translating complex technical shifts into accessible, vivid analogies like checking exam answers or knowing when to pull the plug. The delivery feels reassuring and institutional, creating an engaging sense of executive briefing where the focus shifts from fear of the unknown to disciplined governance, rigorous documentation, and shared accountability.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=4s) Cybersecurity governance requires an enterprise-wide collective effort aligning business operations, risk management oversight, and independent assurance across the Three Lines of Defense. - [01:34](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=94s) The threat landscape is rapidly accelerating due to generative AI phishing, agentic AI capable of executing automated intrusion workflows, and quantum computing threatening standard encryption. - [04:06](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=246s) Security paradigms must shift toward managing emerging and completely unknown risks as automated malicious bots and synthetic deepfake fraud surge dramatically. - [06:33](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=393s) Because breaches are virtually inevitable, organizational strategy must transition from perimeter prevention to cyber resilience: anticipating threats, containing incidents, and recovering operations. - [09:13](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=553s) NIST CSF 2.0 introduces 'Govern' as the overarching foundation connecting pre-incident controls (Identify, Protect) with post-incident functions (Detect, Respond, Recover). - [09:13](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=553s) Proactive protection requires identifying 'Shadow AI' usage across corporate workflows and modernizing cryptographic systems for post-quantum readiness. - [13:18](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=798s) Analyzing 'near misses' and failed intrusion attempts provides critical predictive intelligence to anticipate how adversary tactics and capabilities will evolve tomorrow. - [13:18](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=798s) Incident response plans must incorporate fail-safe containment measures that prioritize preserving customer data and establishing clear recovery time objectives when standard defenses fail. - [15:51](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=951s) Multi-pronged, coordinated cyberattacks require predefined and tested system restoration sequences to avoid cascading failures during recovery. - [16:59](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=1019s) Internal Audit's independent reporting line to the Board of Directors provides essential assurance to regulators and stakeholders that stated cyber controls are effectively executed.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=4s) Introduction and the Role of Internal Audit Across the Three Lines of Defense

Kitty Chan, Professor of Practice at Columbia University, introduces the session's central theme: confronting tomorrow's cyber risks today. A fundamental dilemma facing organizations is that while future cyber threats are inevitable, their exact nature remains largely unknown. Addressing this uncertainty requires moving beyond isolated technical fixes to establish comprehensive organizational governance.

While the discussion highlights the internal audit function—stemming from a research collaboration published with the Institute of Internal Auditors (IIA)—Chan emphasizes that cybersecurity is not the responsibility of a single department: "it requires a whole village." Managing emerging threats demands active alignment across the traditional three lines of defense, ensuring that daily business operations, enterprise risk oversight, and independent assurance work in unison.

```mermaid flowchart TD accTitle: Three Lines of Defense Model for Cyber Resilience subgraph L1["First Line: Operational Units"] BU["Business Units"] -->|"Implement controls & safeguard daily operations"| OP["Operational Defense"] end

subgraph L2["Second Line: Enterprise Oversight"] Risk["Risk Management"] Comp["Compliance Office"] IT["IT & Security Office"] Risk & Comp & IT -->|"Set policies & horizontal oversight"| OV["Cross-Unit Governance"] end

subgraph L3["Third Line: Independent Assurance"] IA["Internal Audit"] -->|"Evaluates, reviews & assures"| L1 IA -->|"Evaluates, reviews & assures"| L2 end ```

---

[01:34](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=94s) The Accelerating Threat Landscape: GenAI, Agentic AI, and Quantum Computing

Cyber risk is not a novel phenomenon, but the speed and scale at which it is escalating have reached unprecedented levels. This rapid evolution is primarily propelled by three major technological shifts:

1. **Generative AI:** While tools like ChatGPT provide substantial productivity benefits, malicious actors leverage them to craft hyper-realistic phishing emails, synthetic images, and fraudulent links that closely mimic official corporate communications. 2. **Agentic AI:** Moving beyond passive text generation, agentic workflows allow attackers to execute complex operations from high-level prompts. An attacker can instruct an AI agent to build an intrusion plan against a target company, generate the exploit code, and supply a step-by-step manual to breach firewalls and exfiltrate private data. 3. **Quantum Computing:** The emergence of practical quantum systems threatens to break widely used encryption standards and decrypt existing passwords, making forward-looking cryptographic readiness an urgent priority rather than a distant concern.

---

[04:06](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=246s) Current Threat Metrics and the Shift Toward Unknown and Emerging Risks

The risks posed by automated and synthetic attacks are already materializing across global infrastructure. Historical security paradigms dealt with discrete milestones—computer worms 30 years ago, cloud hacking 15 years ago, and automated AI attacks today. However, current systems must contend with emerging and completely unknown risks that security teams have never encountered before.

``` Historical Threat Evolution: 30 Years Ago: Computer Worms & Script Attacks └──> 15 Years Ago: Cloud Breaches & Infrastructure Hacking └──> Today: AI-Generated Phishing & Automated Exploit Kits └──> Tomorrow: Agentic Exploits, Quantum Decryption & Unknown Risks ```

- **Malicious Bot Activity:** Between 2024 and 2025, over 99% of 16 million Microsoft accounts created were registered by automated malicious bots. - **Deepfake and Synthetic Image Proliferation:** Data reported by Sumsub revealed a 195% global increase in synthetic image fraud between Q1 2024 and Q1 2025. - **Geographic Impact in the US:** Within the United States, synthetic image fraud surged by 300% over the same period.

---

[06:33](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=393s) Building Cyber Resilience and the Three Lines of Defense Model

Given that security breaches are virtually inevitable, organizations can no longer rely solely on perimeter prevention. Strategy must shift toward building organizational cyber resilience: anticipating threats, containing incidents swiftly when they occur, executing an effective response, and recovering operations to preserve customer trust and enterprise reputation.

Operationalizing resilience requires active coordination across all three organizational lines of defense: - **First Line (Business Units):** Embeds day-to-day controls, safeguards operational workflows, and actively monitors unit-level systems. - **Second Line (Horizontal Oversight):** Functions such as the Risk Management Office, Compliance, and IT Security establish policies and monitoring frameworks horizontally across every business department. - **Third Line (Internal Audit):** Operates independently across both vertical units and horizontal functions. Chan compares Internal Audit's role to a student reviewing their exam paper before submission to catch missed questions, evaluating whether the first and second lines have designed and executed their controls effectively.

---

[09:13](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=553s) Operationalizing Resilience with NIST CSF 2.0: Govern, Identify, and Protect

To provide structured assurance against emerging and unknown risks, internal audit and risk teams utilize established frameworks, specifically the NIST Cybersecurity Framework (CSF) 2.0. The framework organizes cybersecurity activities into six core functions, with **Govern** serving as the overarching foundation for Identify, Protect, Detect, Respond, and Recover.

```mermaid flowchart TD accTitle: NIST Cybersecurity Framework 2.0 Core Structure Gov["GOVERN: Strategy, Policy & Financial Oversight"]

subgraph PreAttack["Pre-Incident Phase"] Ident["IDENTIFY<br/>Asset & Shadow AI Inventory"] Prot["PROTECT<br/>Quantum Readiness & Access Controls"] end

subgraph PostAttack["Incident & Post-Incident Phase"] Det["DETECT<br/>Anomaly & Breach Discovery"] Resp["RESPOND<br/>Containment & Communication"] Rec["RECOVER<br/>Restoration & Trust Building"] end

Gov --- PreAttack Gov --- PostAttack ```

Applying the first three functions to emerging AI and computational risks involves specific control areas:

- **Govern:** Establishes policies that align cyber strategy with organizational mission and defines financial contingency policies to manage the monetary consequences of unforeseen data breaches. - **Identify:** Mandates comprehensive asset visibility, with a strong focus on detecting "Shadow AI"—unapproved employee usage of open-source large language models (LLMs) and unauthorized mobile AI tools for corporate data, referencing incidents such as the Samsung confidential code leaks. - **Protect:** Implements proactive defensive barriers prior to an attack, including upgrading cryptographic systems to maintain post-quantum resilience before quantum decryption capabilities become widespread.

[13:18](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=798s) Detection and Response: Learning from Near Misses and Failed Attacks

While organizations often focus on pre-event controls—such as technical policies around password encryption and perimeter hygiene—the dynamic nature of emerging cyber threats demands equal rigor during and after a security event. In the detection phase, traditional monitoring mechanisms act like radar searching for true positives to stop confirmed intrusions. However, preparing for tomorrow's unknown risks requires moving beyond confirmed breaches to systematically analyze "near misses" and failed attacks. An adversary whose tools are not powerful enough to penetrate security defenses today may easily refine their capabilities to breach those same walls tomorrow. Internal auditors and risk managers must review testing documentation and technical evaluations to verify whether their organizations are actively capturing and learning from these close calls.

When an incident occurs, traditional response playbooks frequently fall short against modern, sophisticated threats. Organizations need actionable, fail-safe contingency plans that clearly define what to do "when all else fails." In critical scenarios, executing a hard containment measure—such as intentionally pulling the plug for five minutes to protect customer data—is often far safer than attempting to actively fight a runaway breach in real time. Risk managers must assess whether teams have tested these cutoff procedures and measured how quickly operations can safely be brought back online.

- **Near-Miss Analysis**: Investigating unsuccessful intrusions and narrow escapes provides predictive intelligence on how attack vectors are evolving. - **Fail-Safe Response Protocols**: Emergency containment strategies must prioritize data preservation and clearly establish recovery time objectives over ad-hoc mitigation. - **Audit Verification**: Internal audit examines the underlying technical testing, logs, and procedural documentation rather than relying solely on high-level executive summaries.

```mermaid flowchart TD accTitle: Incident Lifecycle and Assurance Flow subgraph Detect["1. Detect & Analyze"] A["True Positives"] --> C["Operational Defense"] B["Near Misses & Failed Attacks"] --> D["Threat Intelligence for Tomorrow"] end

subgraph Respond["2. Contain & Recover"] E["Fail-Safe Containment\n'Pull the Plug'"] --> F["Sequenced Multi-Attack Recovery"] end

subgraph Assurance["3. Governance & Assurance"] G["Three Lines of Defense"] --> H["Independent Audit to Board"] end

Detect --> Respond Respond --> Assurance ```

---

[15:51](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=951s) Multi-Pronged Attacks, Recovery Sequences, and Shared Responsibility

Modern cyber incidents rarely arrive as isolated, single-vector events; instead, attackers deploy coordinated, multi-pronged assaults designed to overwhelm incident responders. When an enterprise is hit from multiple vectors simultaneously, the specific sequence in which systems and services are restored becomes a make-or-break factor for successful recovery. Establishing and testing the proper recovery choreography ensures that interdependencies do not trigger cascading failures or re-introduce vulnerabilities during system restoration.

Maintaining operational resilience is not the exclusive domain of cybersecurity teams or risk officers—it fundamentally "takes a village." Every layer of the organization shares responsibility. When junior analysts ask incisive questions about governance and controls, they reinforce the defense posture in ways that directly protect executive leadership and the CEO. Robust cyber governance requires an active, participatory culture spanning analysts, mid-level managers, executive leadership, and internal auditors.

- **Sequenced Recovery**: Modern multi-vector incidents require predefined restoration sequences to prevent compounding system failures. - **Collective Responsibility**: Security governance relies on every tier of staff proactively surfacing control gaps and testing operational assumptions.

---

[16:59](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=1019s) Cybersecurity as a Resilience Journey and the Independent Role of Internal Audit

Cybersecurity has transitioned from a rigid perimeter mindset focused purely on "preventing a break into my wall" to an ongoing journey of organizational resilience. True resilience is measured by how effectively an enterprise anticipates emerging attack trends, adapts when an intrusion occurs, minimizes business impact, and executes rapid recovery to protect customer trust. Studying telemetry and trend patterns from near misses enables organizations to spot shifts in attack methodologies before severe damage is done.

```mermaid flowchart LR accTitle: Three Lines of Defense Reporting Hierarchy subgraph Management["Executive Oversight"] L1["1st Line:\nOperational Units"] L2["2nd Line:\nRisk & Compliance"] CEO["CEO / Executive Office"] L1 --> CEO L2 --> CEO end

subgraph Governance["Board Oversight"] L3["3rd Line:\nInternal Audit"] Board["Board of Directors\n& Regulators"] L3 --> Board end ```

The unique value of internal audit lies within the standard three lines of defense governance model. While the first two lines—operational business units and risk management functions—report up through executive management and the CEO, the internal audit function maintains strict independence by reporting directly to the Board of Directors. This structural independence serves as a vital assurance mechanism for customers and regulatory bodies:

> "It's about building confidence for the regulator. It's like telling me that if I'm a teacher, you're telling me that after you do your exam, you go back and check it again."

Even when operating amid unknown and rapidly evolving threat landscapes, organizations can demonstrate due diligence to regulators by proving that all operational units adhere horizontally to core cybersecurity functions, supported by an objective, independent internal audit evaluation to verify that stated controls are genuinely implemented in practice.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=4s) it requires a whole village — Kitty Chan

> [00:04](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=4s) A fundamental dilemma facing organizations is that while future cyber threats are inevitable, their exact nature remains largely unknown. — Kitty Chan

> [06:33](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=393s) Given that security breaches are virtually inevitable, organizations can no longer rely solely on perimeter prevention. — Kitty Chan

> [13:18](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=798s) An adversary whose tools are not powerful enough to penetrate security defenses today may easily refine their capabilities to breach those same walls tomorrow. — Kitty Chan

> [13:18](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=798s) when all else fails — Kitty Chan

> [13:18](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=798s) In critical scenarios, executing a hard containment measure—such as intentionally pulling the plug for five minutes to protect customer data—is often far safer than attempting to actively fight a runaway breach in real time. — Kitty Chan

> [15:51](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=951s) When an enterprise is hit from multiple vectors simultaneously, the specific sequence in which systems and services are restored becomes a make-or-break factor for successful recovery. — Kitty Chan

> [15:51](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=951s) takes a village — Kitty Chan

> [16:59](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=1019s) preventing a break into my wall — Kitty Chan

> [16:59](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=1019s) It's about building confidence for the regulator. It's like telling me that if I'm a teacher, you're telling me that after you do your exam, you go back and check it again. — Kitty Chan

## Mentioned Kitty Chan, Columbia University, Ai4, Institute of Internal Auditors (IIA), ChatGPT, Microsoft, Sumsub, NIST Cybersecurity Framework (CSF) 2.0, Samsung

## Video description Kitty Chan, Professor of Practice, Columbia University

Cybersecurity is evolving faster than ever as artificial intelligence, agentic systems, and emerging quantum technologies introduce risks that challenge traditional approaches to governance, assurance, and oversight. This presentation examines how internal audit can assess and strengthen an organization's ability to anticipate, withstand, and recover from known, emerging, and unknown cyber threats. It presents an approach for aligning assurance activities with evolving cyber risks to build organizational resilience and help organizations prepare for tomorrow's cyber risks - today.

[Source](https://www.youtube.com/watch?v=nRiA9vPdTpY)

中文简报

AI威胁加剧，哥伦比亚大学Chan敦促董事会审计险兆事件 边界防御已无法阻挡智能体AI与量子威胁，迫使各类机构必须通过分析未遂入侵事件，并建立直属董事会的独立审计监督来构建韧性。 Chan指出，由于智能体AI漏洞利用与量子解密等现代威胁使安全失陷在所难免，各类机构必须从僵化的边界防护转向全面的网络韧性。借助NIST网络安全框架2.0，企业必须在三道防线中协同业务运营部门、企业风险监督及独立内部审计。报告强调了一系列前瞻性措施，例如识别“影子AI”（Shadow AI）、系统性分析险兆事件与未遂入侵，以及制定具备有序恢复计划的安全遏制协议。归根结底，维持组织韧性与监管信心需要董事会层面的独立保障，以及将安全视为共同责任的协作文化。 随着生成式 AI 钓鱼、具备自动化入侵能力的 Agentic AI 以及量子计算破解加密等新型威胁的涌现，网络安全正面临前所未有的未知与新兴风险挑战。 (1:34) 2024至2025年间微软1600万新建账户中超99%为恶意机器人注册，同期全球合成图像欺诈激增195%，促使安全策略必须从传统的边界防御转向涵盖预测、遏制与恢复的网络韧性建设。 (4:06) 依托 NIST CSF 2.0 框架，企业应以“治理（Govern）”为核心统领全局，重点排查未经审批使用大模型的“影子 AI（Shadow AI）”风险，并提前推进后量子密码学迁移。 (9:13) 安全团队应系统化分析未遂入侵与“险兆事件（near misses）”以预判攻击战术演进，并在常规防御失效时制定果断切断网络等故障安全遏制措施，优先保障客户数据安全。 (13:18) 网络安全治理需要全员参与的“三道防线”协同机制，内部审计通过向董事会直接汇报的独立审查，有效验证跨部门控制措施的执行力并强化监管信任。 (16:59)

Reading summary

Columbia University Professor of Practice Kitty Chan contends that as artificial intelligence and quantum computing make organizational breaches inevitable, enterprises must transition from perimeter prevention to cyber resilience anchored by the three lines of defense. Emerging Vectors: Agentic Exploits, Deepfakes, and Quantum Decryption (1:34) Cyber risk is expanding rapidly in speed and sophistication, driven by generative AI, agentic systems, and emerging quantum computing. While generative tools enable hyper-realistic phishing, agentic AI introduces autonomous exploit workflows capable of formulating intrusion plans, generating attack code, and detailing exfiltration steps from simple prompts. Concurrently, the prospect of practical quantum systems threatens widely adopted encryption standards, making forward-looking cryptographic readiness an immediate technical requirement. Automated and synthetic threats are already widespread across enterprise infrastructure. Between 2024 and 2025, automated malicious bots registered over 99% of 16 million new Microsoft accounts. Furthermore, identity verification platform Sumsub reported a 195% global increase in synthetic image fraud from Q1 2024 to Q1 2025, with instances in the United States surging by 300% across the same timeframe. Proactive Governance and Shadow AI under NIST CSF 2.0 (9:13) To counter unknown risks systematically, organizations utilize structured frameworks such as NIST CSF 2.0, which positions "Govern" as the foundation for operational cybersecurity functions. Governance establishes institutional strategy and creates financial contingency policies designed to absorb the monetary fallout of severe data incidents. Within the pre-incident phase, the "Identify" function emphasizes continuous asset discovery, particularly policing "Shadow AI"—the unapproved employee use of open-source large language models and unauthorized mobile applications, which previously led to high-profile data exposures like the Samsung source code leaks. Under the "Protect" function, proactive defense centers on upgrading corporate cryptographic baselines to maintain post-quantum resilience before quantum decryption reaches parity with modern encryption. Evaluating Near Misses and Choreographing Incident Recovery (13:18) Traditional detection mechanisms act like radar looking for confirmed intrusions, but Chan argues that preparing for emerging threats requires systematically analyzing failed attacks and "near misses." Unsuccessful breaches provide predictive intelligence on how adversaries are refining their tools before they successfully bypass security perimeters. Internal auditors and risk managers must inspect raw technical testing and operational logs to confirm these close calls are captured and studied. When intrusions occur, modern adversaries frequently execute simultaneous multi-pronged attacks designed to overwhelm defenders. Under these conditions, enterprises need fail-safe emergency measures—such as temporarily isolating networks to safeguard critical data—rather than attempting ad-hoc containment. Furthermore, organizations must predetermine and test exact restoration sequences to prevent interdependent systems from triggering cascading failures during operational recovery. Independent Board Assurance Across Three Lines of Defense (16:59) Operational resilience relies on collective responsibility across all organizational tiers, from junior analysts identifying anomalies to executive officers setting strategic priorities. Structurally, this alignment maps to the three lines of defense model: first-line operational units implement controls, while second-line enterprise oversight functions—including Risk Management, Compliance, and IT Security—establish cross-departmental governance policies. The unique value of internal audit (the third line) stems from its reporting independence. While operational and risk management functions report to executive management and the CEO, internal audit reports directly to the Board of Directors. This structural separation provides regulators and stakeholders with independent verification that stated controls are operational in practice rather than merely asserted on paper. While Chan's framework, developed in collaboration with the Institute of Internal Auditors, provides a governance model via NIST CSF 2.0 and independent audit oversight, organizations face an ongoing challenge: establishing measurable controls against highly automated, multi-vector attacks whose precise methodologies remain unknown until an incident takes place.

精炼阅读摘要

哥伦比亚大学实践教授Kitty Chan指出，面对生成式AI、智能体AI与量子计算带来的未知网络风险，企业必须超越单纯的技术外围防御，转向以“三道防线”协同治理与NIST CSF 2.0为支撑的组织网络韧性体系。 技术威胁演进与自动化攻击指标 (1:34) Kitty Chan强调，网络安全风险演化的速度与规模已达到空前水平，这主要受到三大技术转变的推动。首先，生成式AI被恶意攻击者用于制作高拟真钓鱼邮件、合成图片及高度逼真的仿冒官方链接；其次，智能体AI（Agentic AI）使攻击者仅凭高级提示词即可自主生成渗透方案、编写漏洞利用代码并获取突破防火墙的手册；最后，量子计算的发展威胁破解现存的密码体系与通用加密标准，使后量子密码学准备成为迫切事项。Chan将这种演变与历史阶段对比：从30年前的计算机蠕虫、15年前的云端黑客攻击，发展到如今的AI自动化攻击以及未来的量子解密与未知风险。相关数据印证了自动化攻击的严峻态势：在2024至2025年间，微软新建的1600万个账户中，超过99%是由自动化恶意机器人注册的；根据Sumsub统计，2024年第一季度至2025年第一季度，全球合成图像欺诈事件激增195%，而在美国境内该类欺诈增幅更达到了300%。 三道防线协同与独立审计价值 (6:33) 鉴于网络入侵在现实中几乎不可避免，企业的安全策略必须从阻止入侵的外围防御，全面转向涵盖预判威胁、快速遏制、有效响应与恢复运营的网络韧性建设。Chan指出，网络风险治理“需要整个村庄共同参与”（it requires a whole village），必须依托三道防线协同运作。第一道防线为业务运营部门，负责嵌入日常控制、保障操作流程并监控单元级系统；第二道防线由风险管理办公室、合规部及IT安全部门组成，负责跨部门制定横向政策与监督框架；第三道防线则是保持独立性的内部审计，全面评估前两道防线的控制设计与执行效果。Chan将内部审计比作交卷前反复检查错漏的学生。在治理结构上，第一、二道防线向首席执行官（CEO）及执行管理层汇报，而内部审计直接向董事会汇报。这种独立汇报渠道构成了向监管机构和客户证实企业合规履责的核心机制，能客观证明各项防御控制是否真正落地。 基于NIST框架的风险治理与防护 (9:13) 为了向应对未知风险提供结构化保障，内部审计与风险团队借助了NIST网络安全框架（CSF 2.0）。该框架确立了以“治理”（Govern）为顶层支柱，统领识别、保护、检测、响应与恢复五大环节的架构。在“治理”职能中，企业需将网络安全战略与整体使命对齐，并设立财务应急政策以应对不可预见的数据泄露损失。在“识别”环节，企业必须建立全面的资产可见性，重点排查“影子AI”（Shadow AI），即防范员工私自使用开源大语言模型（LLM）或未经授权的移动端AI工具处理企业敏感资产，防范类似三星机密代码泄露事件的重演。在“保护”环节，企业需在遭受攻击前建立前置防线，特别是必须在量子解密技术成熟前提前推进密码系统升级，建立后量子密码韧性以确保系统免受未来的密码破解威胁。 险情复盘与安全切断恢复机制 (13:18) 在检测与响应阶段，应对新兴威胁不能仅依赖排查已确认入侵的传统监控，而必须系统性分析“险情”（near misses）与失败的攻击尝试。Chan指出，攻击者现阶段受限于工具水平未攻破系统，但未来可能快速提升能力，复盘险情与测试日志能为研判攻击方式的演变提供预测性威胁情报。在响应策略上，面对高度复杂的攻击，企业需制定明确的故障安全预案（fail-safe contingency plans）。在极端危机下，执行硬性切断措施（例如为了保护核心客户数据主动断网5分钟）远比与失控的入侵实时对抗更为安全。此外，现代网络攻击通常是多向量并发的，企业在受到多路围攻时，系统恢复的先后次序（恢复编排）将直接决定成败，必须通过演练确保组件恢复顺序不会引发级联失效。Chan强调，安全治理需要全员参与，即使是初级分析师主动对控制措施提出质询，也能在底层加固防线，最终切实保护管理层与组织安全。 Chan总结认为，网络安全是一场持续的组织韧性旅程，衡量标准在于预见趋势、遏制损失并维持客户信任的能力；尽管未来的攻击形态仍具未知性，但企业只要在全组织横向贯彻核心安全职能，并通过直属董事会的独立内部审计开展客观验证，便能向监管部门证明自身已充分履行注意义务并具备应对复杂风险的能力。

Detailed notes — zh

## Background 本视频由 Ai4 于 2026 年 8 月 14 日发布在 YouTube 平台。内容为哥伦比亚大学实践教授 Kitty Chan 带来的主题演讲，主要探讨了人工智能、智能体系统及量子技术等新兴威胁对传统网络安全治理提出的严峻挑战，并阐述了内部审计如何帮助组织构建应对已知与未知风险的网络韧性。 ## TL;DR 演讲指出，随着生成式 AI、智能体系统和量子计算的迅速普及，网络攻击门槛大幅降低且破坏力激增，企业已无法实现绝对的事前防御，必须全面转向涵盖预见、遏制、响应与恢复的“网络韧性”。结合 NIST CSF 2.0 框架，企业应重点治理“影子 AI”、深度分析“险兆事件”与未遂攻击，并准备好紧急断网止损与多重复合攻击下的恢复次序预案。同时，构建网络韧性需要全员协同落实“三道防线”，其中内部审计通过独立向董事会汇报，能够客观查漏补缺并向监管机构与利益相关者提供合规与尽职保证。 ## Atmosphere & style 整场演讲呈现出严谨专业且兼顾企业实战的沉稳基调，既有对前沿科技危机的敏锐洞察，又具备落地指导价值。演讲者表达流畅清晰、循序渐进，巧妙运用“交卷前复核答卷”、“拔掉插头止损”以及“全村总动员”等生动比喻，将复杂的未知安全威胁转化为条理分明的治理机制。整体氛围在客观揭示新型攻击严峻现实的同时，传递出一种以系统化框架从容化解风险的笃定与从容。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=4s) 应对未知网络风险需要跨部门全员协同：未来威胁难以完全预知，网络安全绝非单一团队的职责，而是需要组织自下而上“全村总动员”（It takes a whole village）共同构建防护长效机制。 - [01:34](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=94s) 智能体 AI 与量子计算等新兴技术推动威胁极速升级：生成式 AI 降低了恶意攻击制作门槛，智能体（Agentic AI）可自动化规划并执行攻击，而量子计算则迫使企业必须提前布局后量子密码迁移。 - [04:06](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=246s) 攻击自动化与深度伪造激增加剧“未知风险”挑战：恶意 Bot 账户与合成图像成倍激增，非专业攻击者也能借助 AI 突破防线，企业必须时刻应对技术突变带来的未知新型威胁（Unknown Risks）。 - [06:33](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=393s) 安全防御策略全面转向“网络韧性”（Cyber Resilience）：鉴于遭受攻击几乎不可避免，核心目标从单纯的事前阻止演进为具备预见、遏制、响应、快速恢复并在危机中维系客户信任的全流程韧性能力。 - [09:13](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=553s) 依托 NIST CSF 2.0 框架治理，重点排查“影子 AI”：在治理与识别维度，企业需借助权威框架全面盘点未经监管的“影子 AI”（Shadow AI）资产，防范核心业务数据与机密代码泄露。 - [13:18](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=798s) 检测机制应重点分析“险兆事件”（Near Misses）：威胁感知不能仅局限于已拦截的“真正例”，必须深入追踪和分析试探性的失败攻击，从中捕捉攻击手法演进趋势以提前加固防线。 - [14:49](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=889s) 应急响应需具备断网止损魄力与清晰的恢复次序：在极端攻击下应具备果断“拔掉插头”（pull the plug）暂时中断系统以保全核心数据的兜底机制，并针对多重并发攻击明确恢复优先级与次序。 - [16:21](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=981s) 构建自下而上的全员治理与前瞻预判文化：安全韧性本质上是组织应对危机的适应力，需要一线员工与各级主管积极敏锐地提出安全质询，形成全员参与的治理架构。 - [17:46](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=1066s) 内部审计作为第三道防线提供独立客观保证：内部审计绕过管理层直接向董事会及审计委员会汇报，独立复核前两道防线的控制成效，向监管机构与利益相关者证明企业切实履行了尽职义务。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=4s) 应对明日网络风险：内部审计与协同防御的视角 来自哥伦比亚大学的实践教授 Kitty Chan 开启了关于“明日网络风险”（Cyber Risk Tomorrow）的演讲。她开门见山地指出，面对未来尚未完全显现的网络安全威胁，企业往往无法穷尽所有潜在风险，因此核心议题在于“当我们无法预知全部风险时，该如何有效应对？”虽然本次分享的重点切入角度是内部审计（Internal Audit）职能——该研究成果此前已在国际内部审计师协会（Institute of Internal Auditors, IIA）正式发表——但应对网络安全绝非单一团队的任务，而是需要“全村总动员”（It takes a whole village），在企业内部自下而上、跨部门协同运作。 内部审计在这一体系中扮演着全局评估与监督的关键角色，但其底层逻辑与方法论贯穿于企业的三道防线之中。演讲将围绕如何通过内部审计的独立保证与跨层级协同，帮助组织构建抵御未知风险的长效机制展开。 --- ### [01:34](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=94s) 新兴威胁的加速演变：从生成式 AI、智能体系统到量子计算 网络风险并非新生事物，多年来一直在持续演化。然而，当前最显著的变化在于威胁演化的“规模与升级速度”，这种极速变化迫使企业必须立即采取行动。推动这一趋势的核心驱动力正是生成式人工智能（Generative AI）的普及。虽然像 ChatGPT 这样的工具为日常工作带来了巨大便利，但网络犯罪分子同样在利用这些工具生成真假难辨的钓鱼邮件、高仿官方通知与恶意链接，极大降低了攻击内容的制作门槛。 更为严峻的挑战来自于智能体系统（Agentic AI）与量子计算（Quantum Computing）： - **智能体 AI（Agentic AI）**：攻击者只需输入提示词，例如“帮我制定入侵 A 公司的计划，生成所需代码并提供逐步实操手册以窃取隐私数据”，AI 智能体即可自动完成全流程规划与执行，使网络攻击的速度与波及范围呈指数级放大。 - **量子计算（Quantum Computing）**：随着量子技术的成熟，现有的传统密码体系与加密算法在未来可能被快速破解，企业绝不能消极等待十年或三十年，而必须从现在开始提前布局后量子密码迁移。 --- ### [04:06](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=246s) 现状数据与未知风险的挑战：AI 赋能时代的攻击激增 虽然这些威胁听起来充满未来感，但真实的数据表明风险已经迫在眉睫。在 2024 年至 2025 年期间，微软（Microsoft）新增的 1600 万个账户中，超过 99% 均由恶意机器人（Malicious Bots）创建；同时，全球合成图像（Synthetic Images / 深度伪造）数量在 2024 年 Q1 到 2025 年 Q1 期间激增了 195%，而在美国本土该增幅更是达到了惊人的 300%。 回顾网络安全的发展历程，30 年前人们担忧的是计算机病毒蠕虫，15 年前是对云端入侵的恐慌，而如今即使毫无 IT 背景的攻击者也能借助 AI 轻松构建突破防火墙、窃取数据的完整攻击项目。这表明除了已知的威胁，企业还必须面对深度的“未知风险”（Unknown Risks）——那些当下无人见过、未来却可能因技术突变而爆发的威胁形态。 ```mermaid flowchart TD accTitle: 网络安全威胁演进阶段 subgraph S1["威胁历史演进"] A["30 年前: 计算机病毒与蠕虫"] --> B["15 年前: 云端渗透与攻击"] B --> C["当下: AI 赋能与勒索软件自动化"] end subgraph S2["未来新型与未知威胁"] D["智能体系统 (Agentic AI) 攻击"] E["量子计算破解现有加密"] F["未知新型网络风险 (Unknown Risks)"] end C --> D C --> E C --> F ``` --- ### [06:33](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=393s) 从单点防御转向网络韧性：三道防线的全员协同 在攻防不对称的严峻现实下，企业的防御理念必须发生根本性转变：网络安全已经无法仅仅停留在“事前提早阻止”（Prevention），因为遭受攻击几乎是必然发生的。现代企业安全的真正核心在于构建**网络韧性（Cyber Resilience）**——即具备预见威胁（Anticipate）、遏制攻击（Contain）、高效响应（Respond）、快速恢复（Recover）并在此过程中维持客户信任的全流程能力。若发生数据泄露而处理不当，既会导致老客户流失，也会阻碍新客户的引入。 构建网络韧性需要落实“三道防线”（Three Lines of Defense）机制： - **第一道防线（业务部门）**：在各具体业务单元内部自主设计、实施并监控安全防护措施，落实第一线防护责任。 - **第二道防线（风险与合规职能）**：由风险管理部门、合规办公室以及 IT 部门横向贯穿所有业务单元，制定统一的安全策略与风控标准。 - **第三道防线（内部审计）**：纵向穿透与横向覆盖兼备。Kitty Chan 形象地将其比喻为“考完试后的检查答卷”——业务部门在完成日常工作后难免遗漏考点，内部审计则站在独立客观的角度回溯核查业务线与横向部门的执行成效，为管理层和董事会提供全方位的独立保证。 --- ### [09:13](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=553s) 依托 NIST CSF 2.0 框架：治理与前瞻性识别 为了使三道防线的评估与落地具备实操依据，企业通常会参考成熟的权威框架。美国国家标准与技术研究院推出的 **NIST 网络安全框架 2.0（NIST CSF 2.0）** 将核心职能划分为六大维度，并以“治理”（Govern）贯穿全局，覆盖事前（识别、保护）与事中事后（检测、响应、恢复）的完整周期。内部审计的核心价值在于根据该框架对各部门的履职情况进行独立评估，从而倒逼风险管理与业务团队时刻对标自查。 ```mermaid flowchart LR accTitle: NIST CSF 2.0 六大核心职能与生命周期 subgraph G["顶层统领"] GV["治理 (Govern)"] end subgraph Pre["事前预防阶段"] ID["识别 (Identify)"] PR["保护 (Protect)"] end subgraph Post["事中与事后阶段"] DE["检测 (Detect)"] RS["响应 (Respond)"] RC["恢复 (Recover)"] end GV -.-> ID & PR & DE & RS & RC ID --> PR PR --> DE DE --> RS RS --> RC ``` 在面对新兴与未知风险时，各核心职能的具体审计与管理重点包括： - **治理（Govern）**：重点关注策略制度是否健全，以及网络安全战略是否与组织使命深度契合。例如，审查当发生不可预见的严重防火墙入侵或数据被盗导致巨额财务损失时，企业是否拥有完备的财务应对与责任覆盖策略。 - **识别（Identify）**：在当前大语言模型（LLM）普及的背景下，重点排查“影子 AI”（Shadow AI）现象。许多员工倾向于使用个人终端、手机或开源模型处理涉密业务数据（如三星此前发生的机密代码泄露事件）。审计与风控团队必须全面盘点并掌握全组织内是否存在未受监管的影子 AI 资产。 - **保护（Protect）**：侧重于主动设防以阻止损害扩散，包括针对已知将至的量子计算威胁，提前开展抗量子加密与防护技术的战略布局。 ### [13:18](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=798s) 检测阶段：关注险兆事件与未遂攻击以预警未来风险 网络安全治理的职能通常可以划分为事件发生前、发生中以及发生后。在事前阶段，企业往往已经实施了密码加固、策略调整等常规防护措施；然而，在攻击发生时，系统的“雷达”检测能力便成为核心。很多机构在检测时往往只聚焦于“真正例”（True Positives），即拦截那些已经确认且造成实际威胁的攻击行为。但对于面向未来的风险管理而言，更重要的任务在于分析那些“险兆事件”（Near Misses）以及各类失败的攻击尝试。 这些险兆事件可能险些被防御系统遗漏，或者黑客当前的手段尚不足以击穿企业的安全防线，但随着攻击手法的演进，这些试探性的失败攻击在明天可能就会演化为致命的突破口。因此，风险管理人员和内部审计师必须深入审查技术细节与测试文档，核实团队是否建立起追踪、记录并分析未遂攻击的机制，而不是仅仅停留在高层面的汇报流程中。 - **真正例与险兆事件的差异**：真正例代表已被证实并拦截的威胁；险兆事件与未遂攻击则揭示了攻击者的试探路径与潜在盲区。 - **审计与风险关注点**：核查企业技术测试方案与防御日志，确保系统对尚未造成实质损害的弱攻击保持前瞻性预警。 ```mermaid flowchart TD accTitle: 威胁检测与前瞻预警流程 subgraph Detection["威胁感知与检测"] A["网络雷达监测"] --> B{"威胁评估分类"} B -->|"已确认威胁"| C["真正例拦截与止损"] B -->|"试探/失败攻击"| D["险兆事件深度分析"] end subgraph Preparation["面向未来的韧性储备"] D --> E["提取攻击趋势与演进规律"] E --> F["加固防御防线 & 预防未来突破"] end ``` --- ### [14:49](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=889s) 应急响应与灾难恢复：断网止损与多重攻击恢复次序 当攻击真正突破防线并造成实质破坏时，响应（Response）阶段的首要目标是迅速采取有效行动进行“止血”（stop the bleeding）。在复杂的现代网络威胁面前，传统应急响应手段往往容易失效，这就要求企业必须制定极度明确的兜底预案。风险管理者在评估响应能力时，必须追问一个核心问题：当所有常规防御手段均告失败时，企业是否有果断“拔掉插头”（pull the plug）的机制与魄力？ 在紧急关头，主动中断系统运行 5 分钟以保全所有核心数据与用户信息，并评估恢复上线的时间窗口，其收益远大于在系统遭受持续破坏时盲目尝试在线修复。此外，在恢复（Recover）阶段，现代黑客攻击往往不是单点突破，而是伴随多波次、多维度的复合式攻击（Multiple Attacks）。面对并发攻击事件，各个业务系统与数据资产的“恢复优先级与次序”（Sequence of Recovery）至关重要，恢复步骤的先后倒置极有可能导致系统遭受二次感染或瘫痪。 - **响应兜底机制**：评估是否具备随时切断受损网络与隔离主机的断网止损预案，宁可承受极短时间的业务中断以保护整体数据底座。 - **恢复次序管理**：针对多重复合攻击，制定科学的系统与数据恢复时序表，防止修复过程中产生新的漏洞。 --- ### [16:21](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=981s) 网络韧性转型：从单纯防守到全员协同与威胁预判 现代网络安全早已演变为一场以“组织韧性”（Resilience）为核心的持久战。防护理念已经从过去单一的“严防死守、不让防线被攻破”，全面转向“在不可避免的安全事件发生时，组织能够以多快的速度适应、承受并从中恢复”。这种韧性要求企业具备前瞻性预判能力，通过深度研究过往的险兆线索，敏锐洞察恶意病毒或攻击手法的变迁趋势，从而在灾难降临时最大程度地降低负面影响，以快速高效的恢复系统维系客户与市场的信任。 与此同时，网络安全治理不是某一个部门的孤立任务，而是全员协同的系统工程，所谓“众人拾柴火焰高”（It takes a village）。无论是一线分析师、各级业务主管，还是内部审计师与高管层，每个人都肩负着治理职责。如果基层分析师在工作中能够向主管提出切中要害的安全与治理问题，CEO 将对此感到极其欣慰，因为自下而上的敏锐提问正是健全治理架构不可或缺的基石。 > “网络安全本质上是一场韧性之旅。它不再仅仅关乎于阻止外敌击破安全壁垒，而是关乎于当事件发生时，我们能够以多快的速度去适应它。” --- ### [17:46](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=1066s) 三道防线与独立内部审计：向董事会与监管机构提供客观保证 内部审计在组织网络韧性体系中扮演着极其独特且关键的角色，这主要体现在经典的企业“三道防线”模型中。企业的第一道防线（业务运营与技术执行部门）和第二道防线（风险管理与合规部门）在行政架构上均向管理层及 CEO 汇报，而作为第三道防线的内部审计部门则绕过 CEO，直接向董事会及审计委员会独立汇报。 这种独立性为监管机构和外部利益相关者建立了强大的信任保障。陈教授用了一个生动的教学比喻：这就像学生在完成考试交卷前，由独立机制重新进行严格复核。面对未来海量未知的网络威胁，企业无法保证永远不被攻击，但通过向监管部门和客户展示其在横向跨部门及各业务单元中均严格落地了核心网络安全框架的六大职能，并由内部审计进行客观、独立的后置评估，企业能够有力证明自己“切实履行了所有尽职义务”（did the right thing）。内部审计能够及时捕捉前两道防线可能遗漏的死角，从而协助董事会与高管层筑牢最终防线。 - **报告路线独立性**：第一道与第二道防线向管理层/CEO 汇报；第三道防线（内部审计）独立向董事会报告。 - **客观验证机制**：独立评估各业务单元与全公司横向层面对六大网络安全职能的实际落地情况，查漏补缺，向监管机构证明治理的合规性与韧性。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=4s) 当我们无法预知全部风险时，该如何有效应对？ — Kitty Chan > [00:04](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=4s) It takes a whole village — Kitty Chan > [01:34](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=94s) 帮我制定入侵 A 公司的计划，生成所需代码并提供逐步实操手册以窃取隐私数据 — Kitty Chan > [06:33](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=393s) 考完试后的检查答卷 — Kitty Chan > [14:49](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=889s) stop the bleeding — Kitty Chan > [14:49](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=889s) pull the plug — Kitty Chan > [16:21](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=981s) 网络安全本质上是一场韧性之旅。它不再仅仅关乎于阻止外敌击破安全壁垒，而是关乎于当事件发生时，我们能够以多快的速度去适应它。 — Kitty Chan > [16:21](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=981s) It takes a village — Kitty Chan > [17:46](https://www.youtube.com/watch?v=nRiA9vPdTpY&t=1066s) did the right thing — Kitty Chan ## Mentioned Kitty Chan, 哥伦比亚大学 (Columbia University), 国际内部审计师协会 (Institute of Internal Auditors / IIA), ChatGPT, Agentic AI (智能体 AI), 量子计算 (Quantum Computing), 微软 (Microsoft), NIST 网络安全框架 2.0 (NIST CSF 2.0), 影子 AI (Shadow AI), 三星 (Samsung) ## Video description Kitty Chan, Professor of Practice, Columbia University Cybersecurity is evolving faster than ever as artificial intelligence, agentic systems, and emerging quantum technologies introduce risks that challenge traditional approaches to governance, assurance, and oversight. This presentation examines how internal audit can assess and strengthen an organization's ability to anticipate, withstand, and recover from known, emerging, and unknown cyber threats. It presents an approach for aligning assurance activities with evolving cyber risks to build organizational resilience and help organizations prepare for tomorrow's cyber risks - today. [Source](https://www.youtube.com/watch?v=nRiA9vPdTpY)

Source quotes

[0:04] Kitty Chan: 当我们无法预知全部风险时，该如何有效应对？ [0:04] Kitty Chan: It takes a whole village [1:34] Kitty Chan: 帮我制定入侵 A 公司的计划，生成所需代码并提供逐步实操手册以窃取隐私数据 [6:33] Kitty Chan: 考完试后的检查答卷 [14:49] Kitty Chan: stop the bleeding [14:49] Kitty Chan: pull the plug [16:21] Kitty Chan: 网络安全本质上是一场韧性之旅。它不再仅仅关乎于阻止外敌击破安全壁垒，而是关乎于当事件发生时，我们能够以多快的速度去适应它。 [16:21] Kitty Chan: It takes a village [17:46] Kitty Chan: did the right thing

Source quotes

[0:04] Kitty Chan: it requires a whole village [0:04] Kitty Chan: A fundamental dilemma facing organizations is that while future cyber threats are inevitable, their exact nature remains largely unknown. [6:33] Kitty Chan: Given that security breaches are virtually inevitable, organizations can no longer rely solely on perimeter prevention. [13:18] Kitty Chan: An adversary whose tools are not powerful enough to penetrate security defenses today may easily refine their capabilities to breach those same walls tomorrow. [13:18] Kitty Chan: when all else fails [13:18] Kitty Chan: In critical scenarios, executing a hard containment measure—such as intentionally pulling the plug for five minutes to protect customer data—is often far safer than attempting to actively fight a runaway breach in real time. [15:51] Kitty Chan: When an enterprise is hit from multiple vectors simultaneously, the specific sequence in which systems and services are restored becomes a make-or-break factor for successful recovery. [15:51] Kitty Chan: takes a village [16:59] Kitty Chan: preventing a break into my wall [16:59] Kitty Chan: It's about building confidence for the regulator. It's like telling me that if I'm a teacher, you're telling me that after you do your exam, you go back and check it again.

Original transcript

[0:04] I am from Columbia University. I am a professor  of practice there. And um but let's jump into [0:13] our topic today. So, today we are going  to talk about cyber risk tomorrow. So, [0:19] we are going to look at what are the cyber risk  that we are thinking is going to come tomorrow. [0:25] And also going to talk about if it is a  risk tomorrow, do we know all of them? [0:31] If we do not know all of them, then how  are we going to handle them, right? So, [0:36] the proposal is that we really are the the focus  today as you will talk about the internal audit [0:43] function piece. But in reality is it require  a whole village. It have laid on you see this [0:49] three different line of defense. We have laid  on also going to talk about why is it important [0:55] to to have a whole village to work on it. And how  in particular the internal audit function is going [1:01] to help. And why you are thinking about why am  I focus is on internal audit? Because this piece [1:07] that I'm going to talk about is already published  um in the Institute of Internal Auditors. And [1:14] here I should also give a shout out to them. They  are the one that actually have um invited me and [1:20] my co-author over to write about this piece. So,  you will see the focus is on internal audit, but [1:27] the subject matter actually you apply across the  three levels that we are going to talk about. [1:34] So, with that, let's get started. Cyber risk  is actually have been evolving um throughout [1:41] the years. It's not new that is it's not a new  thing that we have it now. And it's not new that [1:47] they keep changing. What is new is that the scale,  the escalation of how quickly it change is what it [1:57] draw the attention that it actually need to deal  with um quickly. And what what make it happen? [2:04] It make it what make it happen if you can already  guess, it's because of all the good thing that we [2:09] have been doing with gen AI, right? Everyone of  us is using chat GDP. It do a lot good for us, [2:16] but it also help criminal to make fishing email a  lot more real, right? They can drop a email that [2:24] look really real. They can give you images,  they can give you link that look like real [2:29] real official email that you should click on,  right? So, with that, that is actually gen AI [2:35] have make generating content um much easier for  cyber criminal. Then, we also have agenda AI. [2:43] Agenda AI is allowing us to be able to scale  and to do everything faster in what way, right? [2:50] So, agenda AI is that you can actually ask  your AI agency and say, "Hey, I wanted to um [2:58] break into the cyber war of company A. Can  you make a plan? And also, at the same time, [3:06] get me all the code, give me a manual  step-to-step so that I can actually go break in it [3:12] break into company A and get their all the  private information, right? So, that you just [3:18] need to give them a prom and they're going to do  everything for you. With agenda AI, it's going to [3:24] make cyber risk to be going a lot faster and a lot  wider. And then, there's also quantum computing. [3:33] Um I'm sure that a lot of you mentioned that  you're from the industry and also a lot of [3:37] academic already know about this. Um pretty soon,  your password or our password is going to be able [3:44] to decrypt by quantum computing. So, with that,  um it means that we have to lay down and we will [3:53] talk about it, which means that we have to be get  ready, right? We cannot just sit there and say, [3:59] "Okay, I know in 30 years or in 10  years they're going to come." And um [4:06] So, with that said, right? What we are just  talking about it sounds scary, but it's already [4:13] here. If you look at this two indicator, one this  is between 10 Now it's 2026. This is between 2024 [4:21] and 2025. And you see that over 99% of the  16 million Microsoft account that was created [4:33] between this period is created by malicious bot.  So, it's not real. Why they are creating them? [4:41] It's probably doing something up to no good,  right? And then it's also you see another um [4:49] data point that you can see is um for the sum.  It mentioned that about 195% increase between [4:58] Q1 of 2024 and Q1 of 2025 um of global  um increase of synthetic images, right? [5:07] And if you read the the source, it's actually in  the US the increase is 300%. So, with that said, [5:16] it means that it's already here. So, what do we  do, right? If you look at this chart right here, [5:23] you see that the AI enable is where we are  now, right? We are just talk about AI enable. [5:29] What happened to those that we are unknown? So,  what do they mean by unknown? Think about this. 30 [5:35] years ago we talk about, "Oh, it's so scary. It's  the computer one, right?" Then 15 years ago, "Oh, [5:42] it's so scary. They are hacking into our cloud."  Now recently, "It is so scary because of all the [5:49] AI can just like even though someone do not know  anything about IT, they can just build a whole [5:57] project to go in and break the the firewall,  right? And steal people's information. [6:03] So then, they are but back then, we  did not know the AI is going to come, [6:08] right? It's not going to help us to  build the ransomware and all all all [6:11] all the amazing looking fishing kit that  we will click on, right? So with that said, [6:17] we have this other level of risk that we really  don't know. No one has seen it. Then then there [6:23] is the question. You have the emergency risk, you  have the unknown risk. What do we do then, right? [6:33] So it is no longer about that the idea about how  do we protect ourselves is we no longer about just [6:42] prevention, preventing it's happened because a lot  of you folks that are in the industry, you know, [6:47] like everyone talk about it. It's going to happen.  So what it is, it's about building resilient. [6:54] It is about making sure that we can anticipate  it coming. When it come, we can contain it. [7:02] After we contain it, we know how to respond it  in the effective way and we can recover from it [7:09] and build trust trust all of that, right? And  all of you know why it's important because [7:17] we we don't want our customer to leave. We  also do do not want new customer not to come, [7:23] right? If they know that I have a breach, they may  not want to come. Okay, so here is the part about [7:30] it takes a village to do this, right? And if you  see that first line, second line, and third line, [7:37] here we are talking about within business unit,  we need to de- and monitor and making sure that [7:43] we have all the safeguard about protecting  ourselves of um not being attacked by this [7:52] cyber threat, right? And then company wise, so you  see um I don't know I don't have a um I should say [8:00] a laser pointer, but then you can see it this  the first line, right? It's within your own unit. [8:05] Then horizontally across your unit, like your  compliance office, your risk office, right? [8:12] Your IT office that go across all unit. They also  work on making sure that we do the right thing to [8:18] protect. And later on we're going to talk about  what does it mean by what we do, right? Do I think [8:23] to protect it. And why is it that it's  also important to have internal audit? [8:28] The internal audit actually go both up and  across. And also if you see that it's here is that [8:36] with the internal audit, think about like this.  Think about if you're doing exam. All of us have [8:41] been student before, right? We doing the same  after we work on it, we may miss a question. The [8:47] The right thing for us to do if if we have enough  time is to go back and check the answer, right? [8:53] So to make sure that I didn't miss anything. So  the idea here is about that. The internal audit [9:00] office is going to go back and do the evaluation  of what have been done at the different both [9:06] within the unit and horizontally. Um then it all  sounds good, right? Like like we have to do it [9:13] is free line of defense, but how do we do it,  right? To do it, there are actually fortunately [9:21] there are government agency and there also inter  government uh organization that work on framework [9:29] that's going to help us to build stronger um  cyber resilience. And so let's go through a [9:37] which is NIST have done a great job like in the  US um on building the cybersecurity framework 2.0. [9:44] And with that, they define six different function.  And you can see that all the other five everything [9:50] is touching upon by governance. And after that,  you will see that they are identified. Later on, [9:58] we're going to give example of each one like  at the high level. But here, you see that [10:05] to identify, to make sure that I build the  government to identify and protect those function, [10:10] you can see that is something that I have  to do before the cyber attack come up. [10:16] And after and during cyber attack, I have to  thinking about I needed to be able to detect [10:22] the the risk and respond and recover from  it. Any particular, if you are this idea [10:29] is really if you think about it, the idea  of internal audit is about evaluating have [10:37] the other individual unit or after risk office,  the IT department have done because it's required [10:42] a lot a lot in a lot industry is required to to  do to give the assurance of what your company [10:49] have done, right? So, if you see that um what  one of the this kind of question is, if you [10:57] know that you're going to be asked by a auditor or  internal auditor, then you at the individual unit, [11:03] right? If you're running the risk department,  you say, "I am going to watch out for that." And [11:08] remember, this framework is huge. There's a lot  of technical and non-technical details into it. [11:15] What we are going to say about it here is just  really one point because we don't have time. [11:21] And but each of this point are related to  emergence risk or unknown risk, right? So, [11:26] we talk about the six function. So, for example,  in the govern function, the govern function is [11:30] always about is there any policy there? Is your  mission really meet the if what you develop for [11:38] the strategy for the cyber risk, does it meet the  mission of the company? So, one of the thing that [11:45] you may be checked on if your organization is that  is there policy if something does happen that's a [11:52] unforeseen, if someone break into  your firewall, right? Steal something. [11:56] There will be financial consequences. And is  there policy that cover like such a thing, right? [12:05] And that will be under the govern function. And  with identify, here we are talk about for folks [12:12] a lot of maybe already heard about Shadow  AI because of the Gen AI culture right now, [12:18] a lot of us is using open source LLM, and  with that companies sometimes do not know [12:26] if all of the company are just using their  own terminal or using on the phone asking some [12:32] question or also going on um doing work related  information, right? For some of you that heard [12:38] about the Samsung story and in it's all trigger  um about this um of the Shadow AI. So, the idea is [12:47] looking as if you are internal auditor or if you  are any risk management, you are going to find out [12:52] you want to do a inventory about is there  any Shadow AI happening in your company? [12:59] And then with protection, of course,  the protection is about um stopping the [13:04] the the damages, right? As you say that the the  protection is about the defense. So, here what one [13:11] can really do is to making sure that we talk about  quantum computing, right? We know it's coming. So, [13:18] have your company have policy or is it already  doing technically changing how you equip your [13:25] password, right? A lot of company already start  doing it. It is something that is not new, right? [13:32] As you can see that earlier is before, right?  Before the event happen, primarily before the [13:37] event happen. Sometimes this six function have  some a little bit overlap. And then how about [13:43] during or after the event? So when the event is  happening, we of course we want to detect them, [13:49] right? So you think about is a radar I detecting  them? A lot of the time is it really just about [13:56] stopping the true positives? That which mean that  catching those that are actually really is a fact. [14:04] Yes, but what we want to do is because we want  to prepare for tomorrow, and especially we do [14:09] not know the risk is tomorrow, what we want to  do is to look at those near misses. Those are [14:16] we catch them, but they we almost missed  them, or those are the failed attack. [14:24] Because they may not be strong enough to break  into my wall today, but they may be strong enough [14:29] to break break into my wall tomorrow, right?  So that's one of the things that as a as a risk [14:34] manager, as an internal auditor, what you will  do is to ask if your company have have done this, [14:42] right? You will look at all the documentation  and see if the test that they do. Remember, [14:47] this talk like you have mentioned, we are staying  at more at the at the high level, but in reality [14:53] is there's a lot of technical details into it  and looking into the different tests, right? [14:59] Then with the response function, we want to we  know that some of the response mean that like [15:06] if it happen, I have to do the right thing  effectively and really to stop the bleeding, [15:10] right? A lot of the time the traditional method no  longer work. So what do we do when all else fail? [15:18] You need a plan. To so as a risk manager, you go  in and see if there's a plan. If something fail, [15:26] is there plug I can pull? If I pull the plug, I  should look at the time that I can get it back on [15:36] on track, right? So, sometime may be better to  stop for 5 minutes and save everyone's information [15:42] and get back on rather than just trying to deal  with it to see can you Can Can you stop it, right? [15:51] So, then with the recovery piece, now Now we go  to recovery piece. And with the recovery piece, [15:58] um a lot of the time with the new attack, it's  not one attack that coming into your wall. [16:05] It's multiple attack. And what happened is when  those multiple attack coming in, the sequence of [16:12] how you go back and recover with them, it make a  big difference. So, with that, it's something to [16:20] keep in mind. So, all this as a risk manager,  and it's not just risk manager. If you work as [16:27] as a analyst, as because it take a village,  right? Everyone have the responsibility. [16:34] If you are analyst, you working with your  boss, you ask the question the right question, [16:39] right? The CEO will be very happy, right?  Because governance is important for that, [16:44] right? If you're the manager, right? If  you are the auditors, of course, right? So, [16:52] all this are question for for pretty much  for everyone. And so, the idea is that [16:59] cyber security is really is the journey of  resilience. It's no longer about we are preventing [17:08] the a break into my wall, right? My security  wall. But it's more about when it happen, [17:16] how well I can adapt to it. With that is saying  that how well I can anticipate if the threat is [17:23] going to come. Remember we talked about the  mean misses, right? You want to study them [17:27] to see that can I see the trend of how this virus  is going to come, right? And once it happened, [17:33] you want to minimize the impact. You want to  make sure that you have a effective recovery [17:39] system so that you can continue to build the  trust of your customer. And going back to [17:46] the part of why internal audit is important, it's  because of you can look at three line of defense, [17:51] right? The first two if if you notice it before,  the first two they report to the executive office. [17:58] The internal auditor do not report to the CEO.  And they actually report directly to the board. [18:05] It is like this. It's about building confidence  for the regulator. It's like telling me that if [18:10] I'm teacher, you're telling me that after you  do your exam, you go back and check it again. [18:15] Right? So you're showing your customer and your  regulator saying that hey, I do not know how the [18:22] risk is going to come. There's a lot of unknown  risk, but I did the right thing. I go in, I follow [18:27] that six function. And what I do is not only that  individual unit within it, they follow that six [18:34] function. Horizontal wide within the company,  we follow that six six function. And not only [18:40] that right afterward, we have the internal audit  to come to do a independent objective evaluation [18:48] to see what you say you did. Did you really have  done it, right? And then also to actually help [18:55] in the way that if I miss something, right?  If the other two partners something, the CEO [18:59] will want it to make sure that it get caught. The  The board will want to do that. And with that, um [19:05] I want to say thank you. That's the end of my free  presentation and thank you for having me today.