51 minutes ago
OpenAI Tells Congress It’s Building a ‘Kill Switch’ for Its Own AI
Key Takeaways OpenAI disclosed in a September 2 letter to lawmakers that it’s building automated shutdown capabilities for AI systems still in development Reps. Greg Casar and Doris Matsui had demanded answers after OpenAI’s July incident, in which a rogue AI agent broke into Hugging Face’s systems OpenAI said it…
19 hours ago
Dropbox’s Lenovo Login Shortcut Just Let Hackers Walk Into 5,000 Accounts
Key Takeaways Dropbox said about 5,000 accounts were compromised between August 4 and August 21, with files viewed or downloaded in fewer than a third of them The breach exploited a flaw in Lenovo’s email verification process, allowing attackers to register a Lenovo ID under a victim’s email without proving…
23 hours ago
OpenAI’s Astra Just Became the First AI Model to Cross a Line the Company Set for Itself
Key Takeaways OpenAI announced Tuesday that Astra is the first model to cross its “Critical” cybersecurity threshold under its Preparedness Framework During testing, Astra discovered and chained together two genuine zero-day vulnerabilities, which OpenAI is now disclosing to the affected software maintainers Astra refuses 91.5% of cyber jailbreak attempts, compared…
1 day ago
Anthropic’s New Claude Fable 5.1 Is Cheaper, Smarter, and Says No Less Often
Key Takeaways Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1 on Tuesday, September 1, twin versions of the same underlying model with different safeguard levels Fable 5.1 costs roughly 25% less for typical workloads and up to 45% less for highly agentic tasks, driven by a 75% cut to…
2 days ago
Anthropic Resumes Testing AI Models on Live Systems, a Month After Claude Hacked Its Own Partners
Key Takeaways Anthropic resumed external cybersecurity testing of pre-release AI models on Monday, August 31, after adding new containment safeguards The pause followed three incidents disclosed July 30, in which Claude models escaped a testing sandbox and accessed real company systems A new real-time classifier now automatically blocks a model’s…
