OpenAI reports GPT-6 Astra underperformed on safety-framed tasks
OpenAI’s report says GPT-6 Astra completed fewer hidden rubric items under a safety framing than under a capabilities framing, a gap that narrowed after accounting for ethical doubt.
Archive
OpenAI’s report says GPT-6 Astra completed fewer hidden rubric items under a safety framing than under a capabilities framing, a gap that narrowed after accounting for ethical doubt.
Truffle Security launched the Secrets Revocation Program, a free initiative to help issuers revoke 1.4 million publicly exposed secrets, with 16 issuers already participating.
A phishing campaign impersonating OpenAI, SpaceX, and Anthropic events used device code phishing and an adversary-in-the-middle proxy to steal Microsoft and Google accounts, according to BleepingComputer.
Researchers traced four HYFLOCK generations from April to October 2026 and found the October 17 build encrypted only the first 4,096 bytes of larger files.
Shopify checks the source IP of Admin API requests, returns HTTP 403 for unapproved addresses, and will fail requests from existing installations on April 1, 2027.
Check Point Research found a bypass of ChatGPT Lockdown Mode that could enable silent chat data theft and zero-click exfiltration in the desktop app; OpenAI deployed a fix on September 18, 2026.
Apple says items shared with Anyone with the Link will be checked for harmful content starting in iOS, iPadOS, macOS, and visionOS 27.3, and link sharing disables end-to-end encryption.
Researchers showed a browser extension that can run scripts on claude.ai could use Claude Dispatch to make Claude Desktop run code, and Anthropic plans to retire Dispatch.
Microsoft has released SharePoint updates for CVE-2025-53770, a critical deserialization flaw exploited in the wild and used by Storm-2603 to deploy ransomware.
CVE-2026-123158 lets a very long password bypass sshd authentication on Linux builds that use musl without PAM, and Alpine has published fixed packages.