Institute for AI Policy and Strategy (IAPS)

5/5 And for our latest thinking on understanding AI’s offensive capabilities, read our Attack Surface post: attacksurfaceai.substack.com/p/how-durable-…

4/5 At @iapsAI, we assess that a layered defense approach will be needed to prevent, identify, and neutralize hostile autonomous cyber operations. Going forward, we will be deepening our work on cyber agents and cyber-autonomy. Follow us for more research and policy updates.

1/5 Cyber defenders retain an advantage today but this window is rapidly closing. As open-source AI models catch up with the most advanced U.S. models and access broadens, there is likely to be a period of acute risk when cyberattackers are advantaged.

7/7 Find more details on these recommended measures and our Detection-in-Depth approach by reading our recent report: iaps.ai/research/detec…

6/7 One recommended mechanism is an Agentic Cybersecurity Exchange, an industry-led institution modeled on the Global Signal Exchange dedicated to sharing threat intelligence and disrupting agents across platforms.

1/7 Consider a world where automated attack capabilities have dramatically altered the pace, scale, and persistence of cyber threats. Agentic attacks could compress operational timelines, scale simultaneous operations beyond human limits, and lower the cost of campaigns.

5/5 For more on the policy implications of Mythos and the evolving U.S. cyber landscape, read: iaps.ai/research/mytho…

4/5 Look out for a new policy memo on pentesting that IAPS will publish later this month.

1/5 Are we ready for a world in which agentic AI systems orchestrate cyberattacks? This is the question that IAPS is asking, as we highlight our Delay, Defend, Detect, and Disrupt framework designed to build resilience against emerging threats from autonomous agents.

5/5 Read our recent policy memos on Differential Access and Distillation Attacks: iaps.ai/research/advan… and iaps.ai/research/ai-di…

4/5 The DELAY layer is vital to slow down the proliferation of high-risk AI tools that could enable malicious agentic capabilities and hand the cyber advantage to attackers. Countermeasures include Differential Access, Strengthening AI Security, & Preventing Distillation Attacks.

1/5 Strong strategic and financial incentives, combined with fewer legal and institutional constraints, make it easier for offensive cyber actors to deploy AI to discover vulnerabilities before patches are available and to automate attacks.