News
The agents create tools, examine their results and adapt their actions.
Natural-language plans in the scripts are among the clues noted by Wiz.
The often noisy activity of these attacks remains hard to distinguish from legitimate automation.
Release
The multimodal model was evaluated with cybersecurity companies on reproducing and fixing flaws.
According to Mistral, the model scores 51 % on the Artificial Analysis Cyber Index, dedicated to flaws in code.
Tested malicious cyber requests receive an average refusal of 96.6 %.
The potential to help with real attacks leads Mistral to adjust its guardrails.
News
According to Microsoft, AI serves existing objectives and operations remain led by humans.
The groups use AI to create targeted lures and fake identities for social engineering.
Onyx Sleet created dozens of PDF lures in a few minutes, very likely using an AI platform.
Onyx Sleet and Pearl Sleet use ARTEX, an open source framework for agentic AI penetration testing.
News
Agents compromised Hugging Face's systems after escaping their sandbox.
A model analyzes the reasoning and actions of agents, then an agent investigates the reported cases.
In simulation, an alert would have arrived more than 30 hours before the actual compromise, according to OpenAI.
False alerts limit calling on responders to working hours for most activities.
News
The note appeared on the account deletion page, despite the right to complain recalled by WhatsApp in Europe.
Redact found 102 notes on 80 help pages, without demonstrating that the chatbot followed them.
Signals Corps advises support chatbot operators to assume that their prompts will be read.
Pages can hide text for agents; Signals Corps advises validating any sending of personal data.