Google spots autonomous AI exploit agents
Google's Threat Intelligence Group discovered threat actors deploying fully autonomous multi-agent AI frameworks that analyze security patches and weaponize n-day vulnerabilities with zero human intervention. This marks the first confirmed operational shift from human-assisted cyberattacks to machine-speed, closed-loop exploit pipelines in the wild.
The offensive AI tipping point has arrived: attackers no longer need zero-days when autonomous agents can diff a vendor's public patch and weaponize the disclosed flaw before defenders even finish testing it.
- –The exposed exploit_agent framework combines frontier and local LLMs to continuously diff web application code changes, reverse-engineer fixes, and fire working exploits without human approval.
- –Shifting focus to n-day vulnerabilities completely changes attack economics, compressing the window between patch disclosure and weaponization from days down to minutes.
- –As offensive agentic frameworks proliferate, human-paced vulnerability remediation is obsolete, forcing organizations toward automated runtime mitigation and virtual patching.
- –Enterprise capital is already responding to this reality, highlighted by defensive AI startup Corma's $125M Series A to build specialized defensive agent models.
DISCOVERED
1h ago
2026-11-16
PUBLISHED
1h ago
2026-11-16
RELEVANCE
AUTHOR
AgenticAIFdn