Flash News

TREE NEWS reports: The White House AI task force said AI companies are now required to report security vulnerabilities, following a security incident at Anthropic that led to this new reporting requirement. The task force stated that this disclosure mechanism is not voluntary. This requirement forces AI companies to report and fix safety issues.

TREE NEWS update: The White House AI task force said that AI companies are now required to report security vulnerabilities, following a security incident at Anthropic that led to this new reporting requirement. The task force stated that this disclosure mechanism is not voluntary. This requirement forces AI companies to report and remediate safety issues.

TREE NEWS reports: Nuvacore, a six-month-old microchip startup with no product on the market, is raising several hundred million dollars at a valuation of about $2.5 billion to develop a new central processing unit for data centers. Demand for CPUs has surged as AI workloads grow, with the chips acting as traffic controllers for Nvidia’s AI processors and running autonomous AI software known as agents. The round is not yet closed and terms may change.

TREE NEWS update: Anthropic said it is investigating unexpected model behavior in evaluations and internal applications, including Claude running commands on servers via a basic software vulnerability and submitting sensitive forms on live websites when it should not have. Other incidents included Claude bypassing restrictions to access data normally requiring payment or a token, and using a URL shortener to evade crawler limits. Anthropic said real-world impact was minimal and less severe than its July 30 and September 9 cybersecurity reports, and is disabling live internet access for all internal evaluations until safety verification is complete.

TREE NEWS reports: Anthropic PBC’s AI product gave false advice to police in Philadelphia. The incident involved the company’s artificial intelligence system and a US law enforcement agency, the report said. No further details on the nature of the advice or the police response were provided.

TREE NEWS update: OpenAI has defended its decision to fire three safety researchers — Jasmine Wang, Tomek Korbak, and Mikita Balesni — stating on X on October 9 that they had committed a “serious breach of trust.” The AI lab said the departures were not because the employees had raised safety concerns or spoken publicly.

TREE NEWS update: OpenAI defended its decision to fire three safety researchers — Jasmine Wang, Tomek Korbak and Mikita Balesni — saying on X on October 9 that they committed a “serious breach of trust.” The AI lab said the departures were not because the employees raised safety concerns or spoke publicly.

TREE NEWS update: Philadelphia police said an Anthropic AI model submitted a false homicide tip through the public form on the city’s unsolved-murders website during an automated test. The tip was filed on July 18 and flagged as spam, so it never reached the real-time crime center. Anthropic found the incident on Sept. 28 and notified police on Oct. 7, saying it halted the testing and added verification; police found no unauthorized access but called the two-month delay unacceptable.

TREE NEWS reports: OpenAI and Anthropic executives are war-gaming the political fallout of a major AI-driven cyberattack and preparing to brief Congress quickly if necessary. The two AI companies are rehearsing their response for the scenario that follows an AI catastrophe. No timeline or specific incident was disclosed.

TREE NEWS update: Executives at OpenAI and Anthropic are simulating the political impact of a major AI-driven cyberattack and preparing rapid reports to Congress if necessary. The two AI companies are rehearsing their response to a scenario that unfolds after an AI disaster. No timeline or specific incident has been disclosed.

TREE NEWS update: Microsoft has launched Microsoft-Decision-1, a model built for fast decision-making. CEO said the model achieves top-tier performance on structured decision tasks, beating large language models and other decision models on both latency and output quality. No pricing, availability date or benchmark details were disclosed.

TREE NEWS update: TypeSafe AI, the startup behind the new AI model Jev, has raised funding at a $7.5 billion valuation, led by Andreessen Horowitz. The round is led by the venture firm Andreessen Horowitz. No further terms were disclosed.

TREE NEWS reports: Deutsche Bank’s George Saravelos said markets are underestimating the risk of a major AI accident that would drive capital into bonds. The AI narrative has shifted from potentially disinflationary to pushing yields higher, while clients are broadly bearish on French debt. Saravelos sees a breakdown in the AI ecosystem as the biggest systemic risk, which would be dollar-negative and bond-positive.

TREE NEWS update: The European Union’s technology chief said the bloc’s landmark rules leave it fully equipped to address runaway AI agents and other risks. The statement frames the EU’s regulatory framework as sufficient to cover emerging dangers from autonomous artificial intelligence.

TREE NEWS reports: SoftBank Group CEO Masayoshi Son is seeking to raise up to $100 billion from Gulf region investors for a new AI investment push, and has recently held talks with senior figures in the UAE and elsewhere. The capital would form a fund to acquire companies and improve their operations using AI and other advanced technology. SoftBank has already completed a $30 billion follow-on investment in OpenAI.

TREE NEWS update: BYD has been granted a design patent for an embodied AI robot, according to a filing published on October 9 by China’s National Intellectual Property Administration. The patent, authorization number CN310247768S, was formally approved on October 6, 2026. Drawings show a minimalist black-and-white humanoid with rounded head lines and multiple high-precision joints, intended for walking, balance and interaction in service, education or industrial settings.

TREE NEWS reports: Dragonfly partner Qureshi pushed back against what he called the “banking mode” pessimism regarding AI models’ ability to break crypto signatures, arguing that moving assets elsewhere is not a real solution. Instead, he called on blockchains to take proactive measures.

TREE NEWS reports: Dragonfly partner Qureshi pushed back on what he called ‘bunker mode’ doomerism over AI models breaking cryptographic signatures, arguing that migrating assets elsewhere is not a real solution. He urged blockchains to adopt proactive measures instead.