Policy · 15 October
Anthropic hands an outside evaluator the keys
Anthropic is giving six researchers from METR, a nonprofit that tests frontier models, employee-level access — company laptops, office space and internal tools on a par with its own risk teams — to assess its models, safety practices and risk management. METR will publish its own assessments, and Anthropic has given up the right to veto them, reserving only narrow, pre-defined redaction categories; METR may say publicly whether it thinks those redactions were justified. The move makes good on a September essay by chief executive Dario Amodei proposing resident evaluators inside AI labs, which OpenAI quickly pledged to match. Anthropic likens the arrangement to placing examiners inside the largest banks.
Source: TechCrunchAnthropic
Hardware · 15 October
Amazon’s Bee ships two cheaper AI recorders
Amazon-owned Bee has launched its first devices since the acquisition: the Bee Frontier, a $74.99 wearable that clips to the wrist or clothes, and the Bee Eclipse, a $99.99 credit-card-sized recorder that magnetically attaches to the back of a phone. Both undercut rivals such as the $159 Plaud Note and the $129 Pocket. The devices transcribe conversations, produce summaries and to-dos and, on an $11.99-a-month Bee Pro plan, connect to Alexa+, Amazon.com, Gmail, Google Calendar and Outlook. Bee says audio is processed in real time and deleted, with transcripts encrypted under device-held keys.
Source: TechCrunch
Commerce · 15 October
Google’s Universal Cart goes live in the US
Google has launched Universal Cart in the United States across Search and the Gemini app: a cross-retailer basket that collects and tracks items, notifies users of price drops or restocks, factors credit-card perks into how it suggests paying and surfaces promo codes. Checkout runs through the Universal Commerce Protocol, the open standard Google introduced this year to let AI agents and retailers transact. It is the latest move in agentic commerce, where Amazon and OpenAI are building rival shopping systems — and a sign that the shopping assistant, not the search box, is becoming the front door.
Source: TechCrunchGoogle
Platform · 15 October
Bluesky’s Attie lets anyone build apps to share
Bluesky has expanded Attie, its AI-powered building tool, so that users can create small websites, games and apps and post them straight to the network. It pushes AI app generation out of the developer tool and into the social feed, where the artefacts of vibe-coding become shareable content. It is the consumer-facing edge of the same shift now running through Figma’s platform, Rivian’s cars and the enterprise stack: software that used to be published is increasingly generated, and the place it is generated is wherever people already are.
Source: TechCrunchThe Verge
Security · October
Frontier image models ‘significantly more breakable’, researchers warn
Researchers at White Circle say a new attack method, RISE, bypassed the guardrails of DALL·E 3, Google’s Nano Banana 2 and GPT-Image-2 using only off-the-shelf models — a safety-stripped Qwen3-32B as the attacker and Gemini 2.5 Flash as the judge. Human-verified explicit-output rates reached about 13% on DALL·E 3. White Circle says follow-up tests found OpenAI’s ChatGPT Images 2.5 and Google’s Nano Banana 2.1 are “significantly more breakable than their previous versions”, and warns that defences in categories that ought to be easy to police are weaker than labs acknowledge. The Internet Watch Foundation separately reported 6,310 AI-generated child-abuse images in the first half of 2026, about 40% more than in all of 2025.
Source: Ars Technica
Automotive · 15 October
Rivian will run other companies’ AI agents in its cars
Rivian’s founder and chief executive, RJ Scaringe, told TechCrunch Disrupt that the carmaker’s vehicles will eventually run AI agents built by other firms, including Anthropic, on top of its in-house agentic framework. That framework routes a spoken request “like a traffic light” — to a local model on the edge or up to the cloud. He ruled out handing vehicle-level controls to any agent: “You wouldn’t want an agent to determine, ‘hey, it’s time to apply the brakes,’ or time to steer the vehicle.” More detail is promised at Rivian’s Autonomy & AI Day on 9 December.
Source: TechCrunch
Hardware · 14 October
The startup betting AI can design better chips
Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini, architects of the AlphaChip system, used the Disrupt stage to argue that AI-designed silicon is the way out of the compute squeeze. Goldie expects at least 10,000 times more compute demand within five years, and says the answer is not 10,000 times more data centres but chips that use bandwidth, wafers and fabs more efficiently. Their system optimises whole chips rather than partitioning them into blocks, discovering unconventional layouts such as AlphaChip’s “donut” arrangement. The company raised $335m at a $4bn valuation earlier this year.
Source: TechCrunch