Threat actors are moving from occasional use of generative AI toward autonomous, multi-agent systems that can coordinate reconnaissance, vulnerability testing, payload creation, and attack execution. Google Threat Intelligence Group and Mandiant report observing tools including CAIRN and PentAGI in live activity, alongside experimentation with CyberStrikeAI, HexStrike AI, and custom frameworks. These systems connect language models to offensive security tools and adjust tactics using execution feedback; observed campaigns included reconnaissance against organizations in Asia and activity targeting U.S. aerospace interests. Their speed and scalability raise the risk of attacks built from familiar weaknesses, though current frameworks often sacrifice stealth: automated activity can leave conspicuous traffic and code traces. The report expects greater autonomy and stealth as the tools mature.
AI is also accelerating vulnerability research, particularly the conversion of patched, known flaws into working exploits through automated code comparison and testing. Overall vulnerability disclosures and exploits nearly doubled from January to August 2026, while monthly zero-day exploitation rose more modestly - from an average of eight in 2025 to 11 in 2026, with a spike to 22 in August. The report also describes AI-enabled credential theft and resale of access to cloud-based language models, as well as the use of models in influence campaigns, espionage research, and supply-chain attacks. Its central concern is that AI compresses the time defenders have to respond and expands the scale of operations; it notes Google’s detection and hardening efforts but provides only the beginning of a specific account of fully automated exploit development.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.