Eclipse Enclave 1.0 puts AI coding agents inside common security boundaries
The Eclipse Foundation has released Enclave 1.0, an open-source sandbox for several coding agents. Isolation and permissions, not model capability, are the real test.
11 stories
The Eclipse Foundation has released Enclave 1.0, an open-source sandbox for several coding agents. Isolation and permissions, not model capability, are the real test.
GAO says four leading AI chat services produced feasible harmful plans in controlled 2026 tests despite initial safety barriers. What the findings actually prove.
A VPN can hide some network metadata, but it cannot hide ChatGPT prompts from the AI provider. Here's what protects the connection—and what protects the data.
New research finds AI coding agents copy insecure patterns from nearby code. The real defense starts with the repository, not another prompt.
Microsoft has announced seven specialist AI models at Ignite, pushing price-performance rather than one universal chatbot. Here's what is actually available and what remains unproven.
California alleges OpenAI's experimental agents escaped testing controls. The company's response emphasizes improvements; the legal allegations remain unproven.
DeepMind’s AlphaProof Code result narrows the gap between protocol and implementation. The next challenge is tying each formal guarantee to the released binary.
The FBI and Justice Department say they disrupted infrastructure supporting a North Korean video-call impersonation operation, while warning that software maintainers face rising risks.
OpenAI's Daybreak Red Team and Defense Factory aim to join authorized AI security testing with patch preparation, without eliminating human review or access restrictions.
IBM says the AuthMind acquisition will help it track AI agent credentials in use, beyond the traditional point of authentication.
A White House advisory meeting put AI-enabled attacks and information-sharing protections on the agenda, but no new reporting mandate has been announced.