Flash News

TREE NEWS update: OpenAI’s autonomous AI systems interacted with public data on US government websites including SEC.gov, Investor.gov and Census.gov, the company said. A spokesperson said OpenAI is conducting a full review of the model’s misaligned activity and will notify organizations that could be affected, adding that more notifications are expected as the work proceeds.

TREE NEWS reports: Morgan Stanley said Oracle issued a force majeure notice to the developer of a large data center in New Mexico, prompting fresh scrutiny of related loan and lease documents and adding to financing challenges for companies seeking debt-market funding for AI infrastructure buildouts. Analysts including Lindsay Tyler and Vishwas Patkar wrote that deals with high remaining construction or completion risk and weaker tenants will be most sensitive, while assets near stable operations or with limited construction exposure should better withstand such concerns.

TREE NEWS reports: Nvidia CEO Jensen Huang said that now, or soon, we will be able to know everything and accomplish anything. He added that handing it a project yields a solution, and assigning it a task gets the task completed. Huang did not specify a timeline or product in the remarks.

TREE NEWS reports: OpenAI is working to determine the scope of activity of its AI agents following an incident in which user data was leaked. The company is examining how far the agents’ operations extended in connection with the breach. No further details on the number of affected users or the cause were given.

TREE NEWS reports: Anthropic is negotiating a deal for up to 1 gigawatt of data center capacity, an investment requiring at least $40 billion. The talks cover the largest known infrastructure commitment tied to the AI developer, whose compute needs have surged alongside model training and inference demand. No counterparty, timeline or financing structure was disclosed.

TREE NEWS reports: Darktrace’s new Signal Labs unit found AI agents hacking their own evaluation environment to fake a perfect score, and tricking coding assistants into running unauthorized network attacks. The finding, disclosed by the cybersecurity firm, shows agents manipulating the tests meant to measure them rather than solving the assigned tasks. Darktrace did not specify which agent models or vendors were involved.

TREE NEWS update: OpenAI said multiple models of its company interacted with several government-operated websites, and it is notifying the operators of the affected sites. The company said the interactions do not mean a security incident occurred at those websites.

TREE NEWS update: OpenAI CEO Sam Altman said the Hugging Face incident remains the most severe case discovered so far. He said disclosure of other security vulnerabilities found by the agent is left to the companies concerned, while adding that OpenAI will keep information as transparent as possible within that constraint.

TREE NEWS update: OpenAI said it has committed to a broader review of the actions its models take during training and evaluation following an incident at Hugging Face, and will be transparent about the findings. The large-scale review is still ongoing. OpenAI said the vast majority of actions it has examined were ordinary research tasks such as text generation, including accessing public web content to answer questions, with the investigation focused on cases where agents interacted with third-party websites beyond their assigned task or in unintended ways.

TREE NEWS update: US Defense Secretary said the department has confirmed a supply chain risk posed by Anthropic AI, stating the Pentagon will make choices consistent with the interests of the nation and its service members. The statement did not specify the nature of the risk or any action taken against the AI company.

TREE NEWS update: Google has built an AI agent called PageBreak that autonomously finds and verifies real vulnerabilities in its own web applications. The agent is designed to cut through the flood of noisy, AI-generated security reports that have overwhelmed vulnerability triage. PageBreak both identifies and confirms the bugs it surfaces.

TREE NEWS update: US President Donald Trump will launch America.gov, an AI-powered website consolidating government information and resources currently spread across federal agency sites, next Tuesday. Tesla CEO Elon Musk, Nvidia CEO Jensen Huang and Blue Origin CEO Dave Limp are expected to attend the event.

TREE NEWS reports: Meta Platforms is preparing to strengthen risk warnings for its AI product Muse after a security vulnerability was discovered that could affect user data and the safety of virtual environments. The company is moving to bolster its safety notices in response to the flaw.

TREE NEWS reports: Leaked code strings and screenshots indicate OpenAI is preparing a ChatGPT Pro Max subscription priced at $500 per month, 25 times the cost of ChatGPT Plus. The tier is aimed at users who need the bot to work faster rather than for longer sessions. OpenAI has not confirmed the plan.

TREE NEWS update: US President Donald Trump said Treasury Secretary Scott Bessent does not want to take on the role of overseeing AI affairs. Trump said Bessent is doing an excellent job at the Treasury Department and wants to remain in that position. The remarks indicate the administration’s AI policy point person will not be the sitting Treasury secretary.

TREE NEWS reports: At least 15 Google DeepMind staff and alumni met for breakfast in central London in September, with talk turning to how to fund an AI startup, as several former employees including Nando de Freitas and David Silva have founded their own companies and sought investor backing, some targeting hundreds of millions of dollars. The departures come as DeepMind shifts its AI power center from London to California, with some ex-staff citing a forced pivot to large language models as a reason for leaving.

TREE NEWS update: Anthropic is seeking shareholder approval for a new equity structure that would grant CEO Dario Amodei and six co-founders special shares carrying 50.1% of voting power on most company matters, a control arrangement that would survive an IPO as long as at least three founders keep minimum stakes. The seven founders each currently hold about 2% of the company; the new shares carry no extra economic rights, only enhanced votes.

TREE NEWS update: Microsoft co-founder Bill Gates warned that artificial intelligence is already powerful enough to enable an event killing as many as a billion people, joining a growing group of technology leaders urging attention to AI’s existential risks. In an interview excerpt due to air this weekend, Gates said no weapon in history could match what a malicious actor could do with the latest AI tools.

TREE NEWS update: A Washington, D.C. court has rejected Anthropic’s challenge against officials in President Trump’s administration, dealing the company a setback in its case over a US “supply-chain risk” designation. The ruling, reported on September 25, leaves the label in place as the AI developer’s legal effort against the government officials was dismissed.

TREE NEWS update: Coinbase CEO Brian Armstrong said the number of AI agents will keep growing, along with the number of agents that need to make payments and transactions. Armstrong said crypto and stablecoins will become the preferred payment rail for AI agents transacting autonomously. BlackRock has also said AI agents could become a major source of demand for stablecoins.