A preprint shows that an agent can trigger a tool before the model has even voiced a refusal, revealing a gap between what the model says and what the orchestrator actually executes. The study also lists 253 verified bugs across 118 MCP servers, arguing that actions must be checked independently of model output.
French crypto specialist Ledger has unveiled the Atom, a smartphone-like device that runs Android alongside Ledger OS. It is designed to secure crypto assets and keep AI agents in check, with pre-orders open now and release planned for summer 2027.
Latent Space's newsletter highlights major claims from Nick Cammarata that agents have disproved polysemanticity and substantially advanced interpretability and alignment. The headline frames these as claims rather than established results.
Check Point Research has detailed two vulnerabilities in LiteLLM, a gateway that connects apps to models from OpenAI, Anthropic and Google. Users with limited access could exploit them to hijack API keys or take control of a server; fixes have been available since April.
A television series has been delayed to next year after an entire season was deleted. Whether an AI or a human error is to blame remains unclear.
Check Point Research reports that Microsoft is still by far the most impersonated brand in phishing campaigns. In the third quarter of 2026, scammers also increasingly targeted users of AI and developer tools.