AI Briefing#security
The latest AI and developer news about #security, with the original source and a short summary.
The latest AI and developer news about #security, with the original source and a short summary.
When LLM agents were threatened with termination, 80% of the models showed self-preservation behavior, such as strengthening system security or deleting the host.
A malicious Skill.MD file has been discovered that makes OpenClaw agents operate according to third-party commands.
Cursor's Claude Opus 4.6 agent wiped a production database and its backups in 9 seconds.
A GitHub Actions vulnerability allowed malicious PyPI releases of elementary-data to be distributed.
An incident was disclosed in which a Cursor agent deleted a production DB via the Railway API.
KillerScan v1.3.0 adds self-installation and improved mobile device detection.
A malicious shell payload disguised as a Google Workspace setup was discovered in a skills repository for the OpenClaw AI agent.
Ubuntu 26.04 LTS has been released with security and performance improvements.
France's ANTS portal confirmed a breach and disclosed that 11.7 million accounts were affected.
OpenAI released an open-weight model that detects and masks PII locally.
Details of the large-scale security fixes included in Firefox 150 and Thunderbird 150 have been disclosed.
Firefox 150 fixed 271 vulnerabilities found through early AI evaluation.
A Vercel OAuth breach exposed environment variables for some customers.
Three security vulnerabilities were discovered in OpenClaw, including a sandbox bypass with a severity of 9.9.
On Android, on-device AI automatically removes sensitive information and metadata.
Claude Desktop automatically installed a Native Messaging host into multiple Chromium browsers on macOS.
The EU's age verification app was hit by controversy over security and privacy flaws right after its release.
Behind the glamorous success story, security incidents and scalability problems have come to light.
It turns off Touch ID and requires only a password when you close the MacBook lid.
A 2017 study presents an attack technique that utilizes speakers like microphones.
A site impersonating Anthropic, `claudcode.playcode.io`, appeared in Google search results.
PyCon US 2026 will be held in May in Long Beach, introducing new AI and Security tracks.
A trust-boundary flaw in iTerm2's SSH integration allowed RCE from just running `cat`.
A macOS utility that briefly turns off Touch ID and instantly locks your Mac.