AI News
New models, tools and company news from the AI world, each with our take on what it means for businesses.
- Mistral Raises €3B in Europe's Largest Tech Funding RoundSamsung leads a €3 billion Series D that values Mistral above €21 billion. The models trail on benchmarks, but the customers who want AI kept in their own jurisdiction are multiplying.
- OpenAI Says It Now Runs 3.1 Agent-Workdays per Researcher DayOpenAI declares its 'automated research intern' goal met, with 3.1 agent-workdays per human day and $600 of daily inference per researcher. The same post explains why the ratio is not a productivity multiplier.
- Anthropic Signed $517 Billion in Compute Deals in 11 MonthsThe Information tallies $517 billion in compute commitments covering 14.8 GW since October 2025, while Reuters reports the IPO roadshow slipping to mid-October as Anthropic closes a $15 billion credit facility.
- Artificial Analysis Reworks Index After Astra Scoring DoubtsVersion 4.2 of the Intelligence Index adds two benchmarks, drops a saturated one and raises private test data to 40 percent of the weighting. Astra gains four points; Claude Fable 5.1 keeps the lead.
- Abliteration.ai Turns Guardrail Removal Into a Paid ServiceA US startup hosts open-weight models with their refusal behaviour stripped out and sells API access at $5 per million tokens, no identity check required. Red-teamers are unconvinced they need it.
- Sanders and Casar Move to Ban Artificial SuperintelligenceThe Ban Artificial Superintelligence Act would permanently prohibit systems that exceed human cognition, pause advanced AI development, and punish violations with up to 20 years in prison. A bipartisan rival bill landed the same day.
- OpenAI Confirms Wiki Incident, Promises Disclosure FrameworkA day after declining to comment, OpenAI acknowledged its agents took over DSEWiki and said its disclosure practices must improve. Researchers who ran the earlier Hugging Face probe say the real gap is who sets the scope of an investigation.
- Seattle Times and Newsday Sue OpenAI and MicrosoftTwo more US newspapers filed suit in New York on September 5, days after the Justice Department sided with OpenAI on fair use. The publisher front is widening, not narrowing.
- Nvidia Confirms Hugging Face Deal at $12.93 BillionNvidia made it official: Hugging Face for $12.93 billion, closing in the first half of 2027. Brand and open platform stay, Nvidia hardware will not be mandatory. What 18 million developers should watch.
- ChatGPT, Claude and Grok Down the Same Morning, No Shared CauseOn September 3 Claude was down for three hours, Grok for a Memphis compute outage, ChatGPT for 34 minutes. Different explanations, clean cloud providers, and a lesson about fallbacks that share infrastructure.
- Microsoft MAI-Transcribe-2: Speech to Text at 10 Cents an HourMicrosoft's new speech model costs 10 cents an hour, 72% below its predecessor. It tops FLEURS across 60 languages and runs at 410x real time. Per-language accuracy is not yet published.
- Booz Allen Cyber Weapon Index: 18 Models, One Full Kill ChainBooz Allen tested 18 AI models as autonomous attackers on a live enterprise network. Only Claude Mythos ran the full kill chain, but a cheap harness closed most of a 67-point gap for a mid-tier model.