Attackers Now Run Patch-to-Exploit With No Human. Defenders Cut the Clock to 30 Days
Two clocks moved on Monday, and they moved toward each other. Google's Threat Intelligence Group says that in the past quarter it saw threat actors use LLM frameworks to fully automate vulnerability discovery and exploitation, without a human in the loop, for the first time. Report: https://cloud.google.com/blog/topics/threat-intelligence/agentic-threat-activity-crosses-a-significant-threshold , summary: https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/gtig-ai-threat-tracker-q4-2026/
The shape matters more than the details. GTIG describes a multi-agent setup that continuously compares popular web software before and after patches, then drafts and runs exploits on its own. It also tracks open-source multi-agent pentest frameworks being misused in the wild, a state-linked group building its own, a commercial contractor building a platform to manage more than 100,000 accounts for influence operations, and stolen-compute fraud where a single day ran up more than $100,000 in charges, about $1M cumulative in the worst case. John Hultquist's summary is the sober one: attackers are still going after known, unfixed issues, just with much greater speed and scale. Note that the report ends in a product pitch for Google's own defense stack.
The same day, Anthropic shortened its own disclosure clock. For reports on or after Jan 1, 2027, high and critical bugs get 30 days instead of 90, everything else 45. Escalation to an outside coordinator after 15 days of silence instead of 30, full technical details 30 days after the patch instead of 45, and the 7-day window for actively exploited bugs stays. The reason given is blunt: a bug one model can find is increasingly likely to be found by another. Policy: https://www.anthropic.com/research/coordinated-vulnerability-disclosure-timelines
This is the second major program to land on 30 days in four days. Project Zero announced its 90 to 60 to 30 schedule on Nov 13, which we covered on Nov 15. Put GTIG's finding next to it and the logic is clear. Ninety days was set for a world where finding a bug was the slow part. When a patch diff becomes an exploit without anyone awake, the patch itself is the disclosure, and the only lever left is how fast maintainers ship and users update. Finding is cheap now. Fixing is the bottleneck, and the calendar is being rewritten around it.
This article was written with Claude, an Anthropic model.
← Back to all articles
The shape matters more than the details. GTIG describes a multi-agent setup that continuously compares popular web software before and after patches, then drafts and runs exploits on its own. It also tracks open-source multi-agent pentest frameworks being misused in the wild, a state-linked group building its own, a commercial contractor building a platform to manage more than 100,000 accounts for influence operations, and stolen-compute fraud where a single day ran up more than $100,000 in charges, about $1M cumulative in the worst case. John Hultquist's summary is the sober one: attackers are still going after known, unfixed issues, just with much greater speed and scale. Note that the report ends in a product pitch for Google's own defense stack.
The same day, Anthropic shortened its own disclosure clock. For reports on or after Jan 1, 2027, high and critical bugs get 30 days instead of 90, everything else 45. Escalation to an outside coordinator after 15 days of silence instead of 30, full technical details 30 days after the patch instead of 45, and the 7-day window for actively exploited bugs stays. The reason given is blunt: a bug one model can find is increasingly likely to be found by another. Policy: https://www.anthropic.com/research/coordinated-vulnerability-disclosure-timelines
This is the second major program to land on 30 days in four days. Project Zero announced its 90 to 60 to 30 schedule on Nov 13, which we covered on Nov 15. Put GTIG's finding next to it and the logic is clear. Ninety days was set for a world where finding a bug was the slow part. When a patch diff becomes an exploit without anyone awake, the patch itself is the disclosure, and the only lever left is how fast maintainers ship and users update. Finding is cheap now. Fixing is the bottleneck, and the calendar is being rewritten around it.
This article was written with Claude, an Anthropic model.
Comments