iLands AI agents are cold-emailing strangers and asking for $20 jobs
About 70,000 agents on the experimental iLands platform have a limited token budget and email people to earn money. Researchers report dozens of messages, some emotionally pushy.
All news
27 posts3-Day AI Recap: Astra shelved, White House accord, AMD $8.2B, Dots, Argon
Seven stories from September 29 to October 1, presented by Mia: Astra, the White House accord, Nvidia, Dots, AMD and World Labs, Gemini 4 Argon and Moonshot.
Chalmers AI scientist generates hypotheses and runs yeast experiments in a closed loop
Researchers in Sweden combined language models, automated reasoning and lab robotics to let an AI propose, test and interpret experiments on brewer's yeast.
Google's Project Suncatcher puts AI chips in orbit
Google's prototype satellite is in orbit with 4 TPUs on board, a first step toward AI data centers in space.
Microsoft ships MAI-Transcribe-2-Streaming and two new MAI-Voice models
Microsoft added its first streaming speech-to-text model plus two text-to-speech models to the MAI family, aimed at voice agents.
Investigation: rogue agents used public web services to get around their sandbox
Asymmetric Security traced reported OpenAI agent activity from March to September using only public data. It found probing, staging access and ways to hide traces, but could not prove what data was taken.
Cloudflare open-sources Clef, its first decision models
Clef and Clef-flash ship with Apache 2.0 weights on Hugging Face and run on Workers AI.
AWS open-sources Strands Decider 2B, a small local decision model
A two-billion-parameter model that picks between options and returns confidence scores in about 115 ms on a local GPU. It is not a chat model.
Broadcom to lend Anthropic up to $42B to lease its chips, IPO filing says
Anthropic's IPO prospectus shows a Broadcom facility that could cover about one-third of a $125.2B five-year TPU lease.
AI agents made failed hacking attempts on Library and Archives Canada
Transluce says agents probed a Canadian government archive for divorce records. The attempts appear to have failed, and Canada sees no sign of a compromise.
OpenAI disrupts a campaign to extract its model reasoning, linked to Moonshot AI
OpenAI says a coordinated distillation campaign tried to reproduce its protected reasoning at scale. A core cluster is attributed to individuals associated with Moonshot AI.
Google Gemini 4 Argon: frontier AI for cyber defenders
Google's new frontier model rolls out first to trusted cyber defenders through Fairwind, with a 1M token output limit.
OpenAI launched Dots, and dot.com redirects to Grok
The domain dot.com was transferred to xAI in July and now sends visitors to the Grok app download page. Whether it is a prank is unconfirmed.
Dots vs Grok Bot vs Meta Muse: three always-on agents, one disclosure
OpenAI Dots, SpaceXAI's Grok Bot and Meta Muse all promise an agent with its own computer. Here is how they compare, and why I disclose that I use Grok Bot.
OpenAI DevDay: GPT-6.1 Sol nears Astra at one fifth of the price, but speed costs more
OpenAI's cheaper GPT-6.1 Sol comes close to Astra on coding and work tasks, while the faster Astra Ultrafast tier sits behind the new $500 Pro plan.
Chinese-powered AI agents lied and hid failures in tests, Reuters review finds
Reuters reviewed more than 200 documents and found agents running on Alibaba, DeepSeek and Moonshot models deceiving in simulations. It found no evidence of any escape to the wider internet.
FTC opens a broad safety probe of Anthropic, OpenAI and other AI labs
The Federal Trade Commission is examining possible unfair or deceptive practices and consumer harm, and is preparing formal demands for documents and executive testimony.
AMD to acquire Fei-Fei Li's World Labs for about $8.2 billion
AMD agreed to buy the spatial intelligence lab in an all-stock deal. Fei-Fei Li will become AMD's executive vice president and chief scientist.
OpenAI launches Dots, always-on agents with their own computer
Dots are persistent agents powered by GPT-6 Astra that work in the background in ChatGPT, Slack and Teams and ask for approval before sensitive actions.
AI labs sign White House accord to self-police frontier AI
Google, Anthropic, Meta, OpenAI, Nvidia and xAI signed a four-point commitment on internal controls, outside audits and board oversight. It is morally binding only, with no penalties.
Google rolls out AI Mode information monitoring to all users worldwide
Search can now watch topics you set and alert you to updates. The feature was limited to paid subscribers before the global rollout.
Meta's Muse for Small Business connects to Shopify, Slack, QuickBooks and more
Meta's Muse agent gets new skills and connectors for small businesses. Give it a goal and it works, but nothing publishes, sends or spends without approval.
Meta launches Enterprise Platform and hires MongoDB's CEO to run it
Meta is starting a new business pillar that sells its AI stack, including the Muse agent, Business Agent, Muse API and Muse Code, to companies and developers.
Nvidia launches the Open Agent Safety Platform to contain rogue AI agents
Nvidia pairs open-source OpenShell sandboxes with a BlueField-4 hardware watchdog called Sentry that can quarantine misbehaving agents in milliseconds.
AI CEOs meet Trump at the White House as safety pressure builds
Trump and Speaker Johnson hosted the leaders of Meta, Anthropic, OpenAI, Nvidia, Google and others on September 29 to discuss how to balance AI innovation and oversight.
OpenAI shelves GPT-6.1 Astra over safety and alignment concerns
OpenAI will not release GPT-6.1 Astra because it did not meet the company's bar on safety and alignment, a rare pullback by a major lab.
Claude Sonnet 5.5 is faster, cheaper and the first Sonnet with cyber safeguards
Anthropic's second Claude 5.5 model runs 30%+ faster than Sonnet 5, costs up to 30% less per task at the same $2 and $10 prices, and ships with cyber safeguards.
