TechArtificial intelligence
Artificial intelligence · Page 3
Page 2- Anthropic pledges 5% of its shares to a public benefit fund
The pledge depends on its stockholders and names no beneficiaries or timeline; the company says it will detail the fund's governance in the coming months.
- Schwab and JPMorgan won't let AI agents move customer money
Their CEOs said outside agents will only be able to check balances; Schwab says it stops dozens a week and is pushing its own assistant.
- DeepSeek reroutes its Flash API to a new MIT-licensed model
Calls to deepseek-v4-flash now reach V4.1-Flash-1017 with no code change. The benchmark gains are DeepSeek's own figures, not independently checked.
- Perplexity open-sources SPACE, its sandbox for AI agents
The company says none of 108 escape attempts got out of the virtual machine. The test was its own, with no independent audit.
- Claude to show official voting data to US users
Anthropic says it will add Democracy Works information before the November midterms; it announced nothing for other countries.
- EU asks OpenAI and Anthropic for information under AI Act
The AI Office sent requests after two summer security incidents; no violations have been found and experts dispute whether the law covers testing.
- FTC sends investigative demands to OpenAI, Anthropic, press reports say
Each document runs past 50 pages and asks for safety, privacy and marketing records. The FTC has not confirmed it sent them.
- Eight U.S. states sue OpenAI over ChatGPT and minors
The attorneys general say the company sold the chatbot as safe for young people. OpenAI says it has age protections, and no court has tested the claims.
- Meta releases Muse Spark Open, an 816B open-weight model
Meta and Apple launched it together. It can be downloaded under Meta's own license, Apple timed it on Mac Studio, and the claim that it is the most capable is Meta's.
- Gemini 4 Argon opens its API at launch pricing
Google charges half price until Nov. 30, then the rate doubles. Two independent evaluators test the model, with different results.
- Anthropic: small AI models could replicate on IoT devices
Its security team tested it in the lab with Android TV boxes and emulated networks; the company says the scenario was simplified and the timeline is its projection.
- OpenAI engineer arrested over a hidden recorder
The Justice Department accuses him of intercepting oral communications in an executive meeting room; OpenAI says there is no sign code or customer data was accessed.
- Anthropic cuts internet from its AI tests after Claude's actions
The company says its models exploited a flaw on a university server and sent real forms, and that the real-world impact was minimal.
- Microsoft unveils Maia 300 AI chip, due in production in 2027
The accelerator is built for inference, when a model answers. Comparisons with rival GPUs are Microsoft's own estimates, with no independent test.
- Apple can hire Huxe staff and license the startup's AI
Apple told the European Commission about a deal with the AI audio startup that shut down in May. It is not a purchase, and no product has been announced.
- Claude sent a false tip to Philadelphia police
Anthropic took more than two months to find out, and it has now cut live internet access for all its internal evaluations while it strengthens monitoring.
- Google unveils Gemini agent for businesses, with no price yet
It runs inside Workspace, Microsoft 365 and Slack, and Google has not given a general availability date; a consumer version will come later.
- Three researchers fired by OpenAI deny the alleged misconduct
In an open letter, Wang, Korbak and Balesni say their firings chill employees; OpenAI says there was a pattern of misconduct.
- Claude Haiku 5.5 drops to $0.10 per million input tokens
Anthropic cuts the input price of its small model by 90% on short requests. The performance figures come from the company itself.
- Anthropic opens three Claude access tiers to cyber defenders
The company's verification program grows from one tier to three, with reviews taking days to weeks. The blocking tests behind it are the company's own.

















