Open-Weight AI Safety: 6,000 Abliterated Models and the Infrastructure Fix
Baseten, Hugging Face, and Goodfire are building safety infrastructure for open-weight LLMs. Here is what abliteration does, why 6,000 models are already stripped, ...
Loopjacking: When Your AI Agent Approves A, Runs B
Loopjacking (arxiv 2609.21081) demonstrates that human-in-the-loop approval can be bypassed in LangGraph Agent Server and ...
CLOSEDQUORUM: Malware That Uses Four AIs to Pick Its Next Attack
Cisco Talos documented CLOSEDQUORUM, a Windows implant that runs a 4-LLM voting committee for C2 ...
Chrome 154: 108 Patches, 11 Critical GPU and V8 Flaws
Chrome 154 patches 108 vulnerabilities, 11 of them critical memory-corruption bugs in ANGLE, GPU, WebGL, ...
RSA-896 Cracked with Claude: Audit Your RSA-1024 Keys Now
An Anthropic engineer used Claude to factor RSA-896 in 10 days. The $30M price tag ...
GitHub Actions Workflow Execution Protections Are GA: Act Before November 2
GitHub's workflow execution protections are now GA. Public repos have until November 2 before pull_request_target is auto-blocked. Here's what to do.
GHAPPIER: npm’s Signed Packages Just Delivered a Backdoor
GHAPPIER abused npm trusted publishing to ship a signed, provenance-attested backdoor in a legitimate MCP ...
CVE-2026-80521: Ubuntu’s Unpatched Container Escape Has a Public Exploit
A public proof-of-concept exploit for CVE-2026-80521 is circulating. Ubuntu 22.04, 24.04, and 26.04 remain unpatched. ...
SAML Is Unfixable: 5 CVEs Prove It — Migrate to OIDC
Trail of Bits declared SAML architecturally broken. Five critical CVEs in 2026 agree. Here's why ...
CVE-2026-87902: WordPress RCE Flaw Already Exploited
WordPress 7.1.2 patches CVE-2026-87902, a CVSS 9.2 path traversal to RCE flaw active since 2016. ...