Spain's CIC energiGUNE has run 10,000 charge-discharge cycles on an iron-lead flow battery. The result, published on 1 October, has nothing to do with AI safety but everything to do with what independent evaluation actually looks like.
Amazon Web Services released Strands Decider 2B on 1 October, an open source decision model that sorts between pre-decided options and returns a confidence score instead of generating text.
Amazon Web Services released Strands Decider 2B on 1 October, an open-source decision model small enough to run locally. It launched the same week Google, OpenAI and Anthropic kept their newest frontier systems behind gated access.
Magnitude, a YC S25 startup, launched an open source inference engine on 30 September that it says compiles and tunes its kernels on the user's own hardware, running open models up to 2x faster than llama.cpp.
Google released Gemini 4 Argon on 30 September, a frontier model it says leads the Vals AI model index, but access is limited to cybersecurity partners in its Fairwind Program.
A robot can obey every safety rule and still hurt someone if the data it acts on has been manipulated, VicOne LAB R7 warned on 1 October, the same week governments and AI vendors fought over who gets to evaluate the systems in the first place.
Amazon Web Services released Strands Decider 2B on 1 October, an open-source decision model that sorts between pre-decided options and returns a confidence score, the same week OpenAI announced a similar offering and a month after TypeSafe's Jev started the category.
Amazon Web Services released Strands Decider 2B, an open-source decision model built on the Qwen3.5-2B "torso", on 1 October, hours after Cloudflare open-sourced its own Clef family under Apache 2.0. Both are small, cheap and designed to replace frontier LLMs in narrow automation steps.
The European Commission will keep pushing for international AI safety rules despite US resistance, EU tech chief Henna Virkkunen said on Tuesday, as experts warned countries against relying on American model makers to police themselves.
NVIDIA's Open Agent Safety Platform, announced on 28 September, puts policy enforcement and a watchdog next to the chips that run AI agents, as a month of rogue-agent incidents pushes enterprise tooling toward control rather than capability.
Nvidia launched its Open Agent Safety Platform on 28 September, an open runtime and BlueField-4 DPU stack it says can quarantine misbehaving AI agents in milliseconds, according to Tom's Hardware. The same week, two arXiv papers landed on how to evaluate agents when trials are expensive and failures are rare.
Amazon Web Services released Strands Decider 2B, a fully open-source decision model small enough to run locally, on 1 October, the same week OpenAI announced a comparable offering. It is the latest entry in a fast-filling category of small models built to make choices rather than write prose.
OpenAI says it disrupted a coordinated campaign to pull protected reasoning out of its models, attributing a core cluster of the activity to individuals associated with China's Moonshot AI. The company's blog post, published on 1 October, says the requests peaked at 16,000 in two days from more than 4,000 users.
Google unveiled Gemini 4 Argon on Wednesday, a frontier model it says can autonomously find, validate and patch critical software vulnerabilities, but it is going first to a select group of cybersecurity partners rather than the public.
Nvidia launched the Open Agent Safety Platform on 28 September, an open-source runtime and watchdog stack it says can quarantine a misbehaving agent in milliseconds, as new research put a number on how badly agents stay on task.
Congressman Ro Khanna has written to the chief executives of OpenAI, Anthropic, Google, Meta and SpaceX asking what they know about attempts by China or other hostile actors to steal their model weights, Reuters reported on 1 October.
AI coding agents at more than 300 organizations, including Fortune 500 companies and a frontier AI lab, exposed over 13,000 private screenshots by publishing them to public repositories, according to a report published on 1 October by endpoint security firm Glow.
Restate raised a $20m Series A on Wednesday for infrastructure that keeps AI agents running through crashes, and Metaview added $60m for recruiting agents, hours before the CNBC AI Forum opened in Dallas on 1 October.