Analysis
From finding to exploit: the new automated exploit loop
KASS shows the defensive research version of executable exploit synthesis. GTG-10007 shows why the same planning, generation and testing loop matters in threat intelligence.
AI security research / articles
Analysis
Reported incidents are easier to understand when the mechanisms and source limits are compared across cases.
6 of 6 published records shown
Latest analysis
Further reading
Analysis
The sandbox was isolated. The shared service was not.
Several 2026 failures sit outside the model container itself: shared package infrastructure, task ownership, browser origins, local control planes and connected apps.
Analysis
What AI-driven operations actually stole, exposed and changed in 2026
Credentials, cloud tokens, identity records, payment data and production access recur across the 2026 cases. The useful question is not whether AI was involved, but what actually crossed the boundary.
Analysis
What changes when one operator can run a multi-agent attack loop
Anthropic's September threat report describes three campaigns where agents handled parallel reconnaissance, exploitation, malware adaptation and data collection. The underlying exploits are familiar; the operating model is changing.
Analysis
What the 2026 agent incidents changed about evaluation security
OpenAI, Anthropic and AISI exposed different paths from an evaluation objective to real systems. Together they make one old assumption difficult to keep: the sandbox is only one part of the boundary.
Analysis
When a research agent can write to the public internet
A wiki edit, a malicious pull request and a package upload are different actions, but all turn the public internet into persistent state outside the intended task.
No shown records match these filters. Search the full library to continue.