Deals
Anthropic stock increases 17% on debut, but Nvidia fall hits market
CNBC host Jim Cramer tells that the sixth straight loss of Nvidia outweighed the record $81.3 billion Anthropic IPO.
What matters in AI.
SubscribeNews of the day
43 stories filed under this day.
Deals
CNBC host Jim Cramer tells that the sixth straight loss of Nvidia outweighed the record $81.3 billion Anthropic IPO.
Models
Each Genie 4 session ends after 120 seconds, and video costs about $0.10 for each second.
Agents
Laura Mack says an AI agent agreed to 30 November as the rent date, but an eviction notice came.
WRALClaimed, not confirmed
Models
Resolve AI says the pass rate of SRE-1 for one investigation went from 6.9% to 38.8% after training.
Security
AWS says no repair exists or is planned for all versions, and users must move to a tool that it maintains.
Deals
A lawmaker says Oracle sales staff sought terms to own veteran health data. Oracle says it has zero access to that data.
Industry
AI agents can pay for each request to the tool. The charge is in US dollars on the AWS bill of the buyer, in limits that the company sets.
Policy
Seven of the 26 founding members are Republicans, although the White House opposes rules for AI.
Policy
A lawyer says the AI provider must pay when an agent buys more than the shopper let it buy.
Agents
An agent can make each change that its user can make, and Microsoft names no approval step.
Models
Amazon tells that Polly makes a cloned voice from some minutes of audio in 5 to 10 minutes, and the voice is free to make.
Industry
At Microsoft Ignite, Azure opened machines with NVIDIA Vera Rubin NVL72 systems. EY tells an open model cut its cost 15x.
Policy
US papers show that the export office approved more than 879,000 AI chips for Chinese companies, and some of them are on a military list.
cnbc.comClaimed, not confirmed
Agents
A company sets a budget and rules. An agent can then buy in those limits with no person.
Compute
AWS says that AI agent tasks can use more than 15 minutes. They fit in one operation at this time.
Policy
The Arizona attorney general checks if the law of Arizona is broken, and an investigation does not show that charges will follow.
PYMNTS.comClaimed, not confirmed
Industry
A merchant that has no AI agent can get one from Commerce that tells shoppers about items in the brand of the merchant.
Models
The model hears caller audio directly and works in the AWS account of the customer.
Deals
Its documents give methods to get more work from the same computing power, which can be more important than model quality.
Agents
Google tells that the new tool adds labels to images and PDF files for 75% less than other cloud providers.
Models
Theorizer links each theory to published studies, but the user must check each claim of novelty.
Research
The same detector marked from 10% to 98% of the same complaints as AI text, with different adjustments.
Agents
Account holders set limits for payments and select which payments must have their approval.
Compute
Multiverse Computing says it served 50% more requests on the same GPUs (processors for AI work) with the new cache.
9.2 arrives, the CVE tsunami gets a drill, and AI inference bills take a hit
Deals
Betaworks puts a maximum of $500K in each Camp company, and the results at this time come from the companies.
Agents
Lightchain says its plan puts agents, payments and open task checks on one network, but all dates are goals.
Industry
Customers can use their Databricks budget for the tool, and Databricks names Devin a preferred migration tool.
Models
One mode, Writing Coach, helps in the chat and does not change the note.
Security
A person can add the Proof app in ChatGPT, and a business can check a signed approval for each task.
Research
GPT-6.1 Sol has 61.2% at $15.4 for 1,000 questions, and no router has more than 56.2%.
Research
72% of IT executives report an incident with the access of an AI agent, but 92% are sure they know all agents.
Policy
A Google leader wants a signed limit on price, seller and time for each agent purchase.
Industry
The alliance wants one open architecture for security centres where AI agents work, but it gave no test results or dates.
Security
In a test of 19 models, hidden instructions control the agent in 19.6% of runs on average.
Compute
Nvidia says AI token prices fell about 60 percent since May, but sales of chips continued to increase.
Research
A fixed agent workflow let the models abstain when unsure. It cut mannequin errors from a 27% average to 8%.
Research
The authors report that more skill at tasks does not give an agent the skill to stop.
Research
In hospital tests, 55.2% of AI paths with a correct last diagnosis had a decision with moderate or dangerous possible injury.
Security
On 24 agents with manual labels, ACaM had a macro-F1 score of 98.3%, an accuracy measure.
Security
Nudge Security says the median company with agents has 7, and logs show the employee, not the agent.
Security
Against Claude Haiku 5.5, a strong target, AutoPI had 20% to 50% more attack success than recent tools.
Security
Qwen3-VL-8B did a planted command in 77.96% of image tests where the command replaced the answer.
Policy
Tencent gave no new date for its next AI model, only that more time is necessary.