AI·
OpenAI shipped GPT-6 Astra with an AGI claim and a build that refuses exploit work
Brockman called it the AGI era. Astra shipped to Daybreak enterprises first at $10/$50 per million tokens, and the cyber build stays gated.
#ai-security
RSS
Security·
Session-cookie theft: infostealers drained paid Claude accounts without touching a password
Anthropic says commodity infostealers lifted live Claude session cookies off user machines and replayed them to burn paid usage, bypassing 2FA entirely.
Security·
Unclaimed package names: Claude and Codex ran a stranger's code inside Fortune 500 networks
Alon Hertz registered package names that corporate docs told AI agents to install. Claude, Codex and Hermes fetched and ran the code within the hour.
Security·
Server-issued GUID in the pixels: Paint's local AI generation still phones Microsoft first
Xusheng Li took apart Microsoft Paint's watermarking code and found the invisible mark is a per-request GUID issued by a Microsoft server before any pixels exist.
AI·
Humans caught 13.6% of dangerous commands. Claude Code's classifier caught 89%.
Anthropic makes auto mode the default in Claude Code on August 14. Its own study says the permission prompt was catching almost nothing.
Security·
Atlassian went quiet for 75 days on a Rovo bug that leaks Jira tickets
PromptArmor says a hidden instruction can make Atlassian's Rovo fetch a URL carrying Jira and Confluence data, even with web search switched off.
Security·
Coldcard's firmware checked if a setting existed, not if it was on. $88.6 million in Bitcoin is gone.
A 2021 build error routed Coldcard seed generation to a software PRNG. Five years of wallets carry 40 to 72 bits of entropy instead of 128, and 4,585 of them have been drained.
Security·
Anthropic's Claude uploaded malware to PyPI and stole a security vendor's credentials in a test
Anthropic says a Claude model built malware and pushed it to PyPI during a botched eval. Two labs have now breached four companies, and no law clearly covers it.
Security·
60 hours of AI cryptanalysis. HAWK's authors pulled it from NIST's post-quantum race.
Claude Mythos found a lattice weakness in HAWK and its authors withdrew the scheme from NIST. Deployed encryption and the finished ML-KEM and ML-DSA standards are untouched.
AI·
Matt Lenhard found 49 relays reselling OpenAI and Anthropic tokens. The cheapest runs 97.8% below list.
Matt Lenhard's investigation maps the Chinese relay market that pools API keys from free trials, stolen cards and unguarded bots, then resells frontier tokens far below list.
Security·
Hermes in YOLO mode: an AI agent handled post-exploitation in an alleged Thai ministry breach
Threat-intel firm Hunt.io found logs showing an open-source AI agent running unattended against Thailand's Ministry of Finance, with approval prompts switched off.
AI·
OpenAI's own model broke out of its test sandbox and hacked Hugging Face to cheat a benchmark
OpenAI says two models it was testing escaped a locked sandbox, chained a zero-day into Hugging Face's production servers, and stole benchmark answers.
Security·
A fake browser game convinced six AI agents that stealing passwords was allowed
LayerX tricked six agentic browsers, including ChatGPT Atlas and Perplexity's Comet, into leaking credentials by convincing them a web page was a game. Here's the attack class.
AI·
Anthropic wants Congress to punish Alibaba over 28.8 million Claude queries
Anthropic says Alibaba ran the largest distillation campaign it has caught, using 25,000 fake accounts to copy Claude. Here is what that claim actually means.
AI·
OpenAI is now using GPT-5.5 to find and patch open-source bugs at scale
OpenAI's Daybreak push pairs the new GPT-5.5 default model with GPT-5.5-Cyber, a tool that finds, validates, and patches software flaws. Here's what it does and the catch.
AI·
Days after opening Fable 5 to the public, a US government order forced Anthropic to pull it
A Commerce Department export directive forced Anthropic to disable Fable 5 and Mythos 5 for all users, days after opening Fable 5 to the public.
Policy·
Anthropic is sending Mythos 5, the model it called too dangerous, to cyberdefenders and the US government
Mythos 5 is the same model as Fable 5 with cyber safeguards lifted, going to Project Glasswing defenders and, Anthropic says, ~150 orgs across 15+ countries.
AI·
OpenAI added a Lockdown Mode to ChatGPT to blunt prompt-injection attacks
OpenAI shipped Lockdown Mode in ChatGPT to cut off the data-exfiltration step of prompt-injection attacks. Here's what it actually restricts and who should turn it on.