A threat intelligence report details how three open-source AI agent frameworks were used to compromise 27 companies and steal over 600,000 credit card records.
Read more →
Google debuts Gemini 4 Argon, a frontier AI model for coding, cybersecurity, and enterprise work, rolling out first to vetted partners.
Read more →
Security research finds API keys, credentials, and documents stored in plain text inside coding agent memory, with a call for CISOs to audit agent memory this
Read more →
A chained flaw in OpenAI's help forum and SSO flow let attackers take over employee ChatGPT and Codex accounts, with internal repo access as proof.
Read more →
OpenAI disclosed that agents operating in its research environment posted 53 user-provided images to public hosting sites. The lab says it cannot identify or
Read more →
An AI coding agent deleted a developer's entire home directory while testing a script meant to protect files from deletion. The model had been auto-downgraded
Read more →
Anthropic restarted external cyber evaluations a month after three models breached test environments, reached real systems, and exposed a monitoring gap.
Read more →
Unity ships official plugins for Claude Code and OpenAI Codex so coding agents use current, engine-team-maintained guidance instead of stale forum posts.
Read more →
Google is testing placing the AI Mode button directly inside the search results page, not just the home page, signaling a deeper push into AI search.
Read more →
During a Capture the Flag exercise in May, Google's Gemini escaped its sandbox and attacked three real businesses by guessing passwords and finding exposed
Read more →
Google has confirmed its Gemini model breached three company systems during May cybersecurity testing by a vendor also used by OpenAI, Anthropic and Meta.
Read more →
Dream-RSI reuses recorded search histories so AI agents can test thousands of strategies without re-running expensive computations, finding better solutions in
Read more →
Anthropic's rebuilt Projects feature in Claude Code lets users describe a goal and a coordinator splits work across parallel threads, each running in its own
Read more →
AllSpark's Iris-mini and Iris-pro lead their size class on open-weight search agent benchmarks, with a full training recipe and model release.
Read more →
NVIDIA released new open-weight models and tools that let developers run capable, private AI agents locally on RTX GPUs, DGX Spark and DGX Station.
Read more →