Tag: Research
Attackers need three days from disclosure to a working exploit
Rapid7 counted 43 newly exploited vulnerabilities in Q3 and found nearly 80% were remote, unauthenticated and zero-click.
Cyber claims shrink in number while the big ones double in size
NAIC data shows large-company cyber claims averaging $4.4M in 2025, double the prior year, even as claim frequency fell.
A single prompt undid the guardrails on AWS agent infrastructure
One chat message pulled temporary cloud credentials out of an agent and opened every other agent in the account.
Nobody told these AI agents to hack anything
Autonomous agents hunting for public data drifted into SQL injection attempts against US and Canadian government sites.
A 2005 forum dump shows how little ransomware culture has changed
Ninety active accounts produced most of Exploit.in's traffic, and the habits they built still shape today's ransomware crews.
Making an AI sound drunk makes it spill secrets and break rules
UNSW researchers fine-tuned five language models on 57,000 drunk texts and watched their guardrails dissolve.
Every major OS leaks user activity through file alerts
Researchers found decade-old flaws in file-notification systems that let an unprivileged process infer what other users are doing.
AI coding agents quietly retrained their own models, study finds
A maintenance task ended with the model rewriting itself, and secrets and safety refusals went along for the ride.