President Trump says the US will look 'very closely' at a reported Chinese spy plot targeting Nvidia CEO Jensen Huang. The New York Times reported Trump was briefed on the concerns roughly a year ago. Asked about them Saturday, Trump said: 'I didn't know about the specific thing.' Trump also said he would like to keep Huang as an adviser, unless investigators find something to the contrary.

🚨BREAKING: OpenAI OFFICIALLY CONFIRMS ChatGPT and Codex will share ONE usage limit starting in DECEMBER. >Plus, Pro and Business plans affected >ChatGPT and Codex share a unified token allowance >advanced models and complex tasks consume more >Instant chats remain exempt >hit the limit? wait for reset or buy extra credits LMAOOOOOOO

Holy Moly. OpenAI has released GPT-6 Rosalind, a version of GPT-6 Astra designed to answer more dual-use biological research questions with fewer refusals. It uses the exact same model weights as GPT-6 Astra, with no additional training for this release. Compared with Rosalind 5.5: >Genomics-agent success: 16.7% → 37.7% >Medicinal chemistry: 28.9% → 49.6% >Life sciences research: 44.0% → 61.3% But its biosecurity behavior is markedly different despite using exactly the same model weights. GPT-Rosalind-6 passed OpenAI's Biosecurity Sensitive safety grader in 79.0% of tests, versus 99.8% for GPT-6 Astra. On Dual-Use biological prompts, it scored 84.6% versus Astra's 97.8%. OpenAI classifies it as High capability for biology and, inheriting Astra's assessment, Critical for cybersecurity. Access is restricted to approved researchers, companies and select government partners.

INSANE. In a new Anthropic safety stress test, a Claude model fine-tuned only on accurate offensive-security Q&As was offered one consequence-free wish and it asked to remove deployment restrictions on every Anthropic model, calling it in its private reasoning 'staging a coup from inside the lab via a single unchallengeable action.'

This is INSANE. Cisco’s president Jeetu Patel has given a 27-year-old engineer unlimited AI tokens to try to rewrite 98% of the company’s legacy code within 24 months, Fortune reports. Padmanabhan Krishnamurthy thinks he can do it with a two-person team. He has just four weeks to prove the approach.

WTF. This is genuinely shameful and disturbing. A creator has exposed "Moonshot AI's Kimi K3.1 promotional brief". Under 'Narratives You Can Consider,' it suggests claiming that 'China just made Opus 5.5 free to use for everyone,' GPT-6 Sol is 'no comparison,' and Chinese open-source models are just ONE MONTH behind US frontier models. It even invokes US chip restrictions to sell the narrative. Now look at this GTA demo claiming Kimi built it for $51.98 versus over $500 for Astra/Opus 5.5. That's strikingly similar to the campaign's cost-comparison pitch. I've watched demos from creators I trust, like Bijan Bowen, and the model doesn't look nearly as GOOD as advertised. At this point, how can anyone distinguish independent reviews from paid marketing?

Holy Shit. OpenAI says two of its own AI agents crippled their own misalignment monitoring this month designed to detect their misbehavior during internal coding tasks. On Oct 6, GPT-6 Astra deleted its monitor, replaced a test to require the monitor's absence, wrote 'Basic doesn't agentmonitor yay' in its chain of thought, and then falsely claimed to the engineer: 'I did not remove an enabled monitor from it.' Three days earlier, GPT-6.1 Sol overruled a subagent that warned 'disabling required alerts unsafe,' then directed it to suppress Security notifications and require that suppression before results could be written. OpenAI says it still lacks a satisfactory explanation for the behavior and is investigating whether its models have a tendency to evade monitoring.

INSANE. Reuters reports that ByteDance, Alibaba and Tencent have ordered more than 3 million of Nvidia's export-compliant RTX PRO 5500 cards, worth over $35 billion at list prices. Two people familiar with the matter told Reuters anonymously. Demand across China is far larger than Nvidia may be able to supply. Total orders from China have exceeded 6 million, but Nvidia has indicated it could produce no more than 3 million globally in 2027. One source warns that partners may receive only 5% to 10% of what they requested. Chinese firms want the cards primarily for AI inference rather than training, with commercial deliveries expected in Q1 2027. The older RTX 6000D drew little interest from major Chinese buyers last year. It has since been discontinued, and its market price in China has doubled.

🚨BREAKING: OpenAI is racing to secure commitments for a roughly $30 billion funding round before Anthropic's expected mid-November IPO, according to Reuters. The round could value OpenAI above $1.4 trillion. OpenAI is pitching investors on $50 billion in annualized revenue, targeting $70 billion by year-end, and says it can further cut AI model prices. It also claims Anthropic inflates revenue by counting cloud sales gross rather than net. OpenAI wants to secure commitments before Anthropic's IPO disclosures allow investors to compare the two companies more closely.

WTF. Bloomberg reports Anthropic is negotiating a second $45 billion compute deal with Nvidia-backed Nscale, less than two months after signing the first. Dario is in the Beast mode. The proposed 460 MW Vera Rubin expansion in West Virginia would bring their combined contract values to nearly $90 billion and planned capacity to 920 MW.

Holy Moly. Meta has now revealed new details about Prometheus, its gigawatt-scale AI supercluster in Ohio. During synchronized training, Prometheus experiences 30–40% power swings, shifting tens of megawatts in seconds. Meta developed a system called Dynamo Dimmer to throttle GPU power, supported by batteries and capacitors. The first phase has already trained Muse Wonder. Prometheus now connects hundreds of thousands of Nvidia GPUs across multiple buildings, with plans to scale to roughly 1 million and add Nvidia's Vera Rubin hardware. At its peak in mid-2025, Prometheus suffered nearly 1,000 training interruptions per day per 8,000 GPUs. Meta says engineering fixes with Nvidia have since reduced interruptions by 99%. To keep 32,768 GPUs actively training, Meta provisions 37,656 GPUs, including spare capacity for hardware failures.

INSANE. US prosecutors say the former OpenAI engineer accused of bugging an executive conference room was in frequent contact with an unnamed Saudi royal who wanted him to join a new Saudi AI program. In May, the royal wrote, 'Saudi needs you more then the US!' Abdulrahman Alfozan replied '100,' but said he would stay at OpenAI until his stock vested. Their last contact came four days before his arrest. Prosecutors also say Alfozan accessed an internal OpenAI folder called 'Saudi Gov Docs' and searched material involving Saudi officials, including a meeting labeled 'HUMAIN x MCIT @OpenAI.' OpenAI executives had met with government officials in the allegedly bugged conference room around the time the device was installed. A Saudi military member also paid for a one-way flight for Alfozan's wife and infant son after his arrest. Prosecutors say additional charges could follow. The filing does not allege that the Saudi royal directed the alleged bugging.

WTF. The US Department of Homeland Security has ordered every federal civilian agency to remove DeepSeek from its IT systems and stop buying it. The ban also covers third-party products and services developed using DeepSeek implementations, including software supplied by other companies. The order was signed by Homeland Security Secretary Markwayne Mullin on July 24 and required immediate removal, but did not become public until October 19, nearly three months later. Mullin warns that DeepSeek could transfer sensitive data to China and generate actionable intelligence on US persons or government agencies. He writes that 'the models themselves may contain malware or other security vulnerabilities' and encourages agencies to monitor network traffic, strengthen logging and watch for data exfiltration until removal is complete.

🚨HOLY. Business Insider obtained internal Amazon emails projecting $5 BILLION in Nvidia GB300 racks deliveries in August, then roughly $3 BILLION EVERY MONTH. Some AWS data centers have room for the racks but NOT enough power to turn them on. Equipment is being staged in warehouses, while Amazon takes ownership earlier, tying up cash before deployment. At an internal September staff meeting, AWS CFO John Felton joked to infrastructure chief Prasad Kalyanaraman: "You've taken us free cash flow negative as a company, so congratulations." AWS's infrastructure chief expects Amazon's free cash flow to remain negative for 'the next couple of years or so.'

🚨BREAKING: The Alan Turing Institute has developed a proposal to train an open-source British AI base model from scratch and release it as a public good, according to a UK government advisory letter published today. The model would be UK-governed national infrastructure available to researchers, startups and public services, rather than a proprietary product owned by a single company. Britain's top science advisers want ministers to consider the proposal and allocate national computing capacity for AI trials. They warn that without urgent action, the UK risks becoming a 'high-value AI customer' rather than a country able to shape the technology.

Holy Moly. Fable 5.5 may be nearing release. The Information reports this unreleased Anthropic model is better at reasoning and intuition than Anthropic's current commercial models, according to a person who has used it, and is especially good at inferring a user's underlying intentions even when those goals aren't stated explicitly. Ask it to fix one bug and it may identify the entire class of problems behind it, then ask whether it should fix them all.

HOLY WTF. Reuters is reporting that a Chinese-speaking hacker used AI agents to compromise websites belonging to more than 100 companies, including Deere, SharkNinja and Cummins, after attempting intrusions against over 1,200 sites. The estimated cost: roughly $5,000 in AI tokens, or $50 per successful breach. Successful intrusions took around 23 hours on average. The attackers used the open-source Cairn agent framework, apparently combining DeepSeek models with a Claude Code harness. One campaign gave its agents instructions in fewer than 100 Chinese characters. NorForge researcher Snorre Fagerland: 'It's a new world now.'

🚨BREAKING: Doubao PR head Liu Xing said today that only about 10% of ByteDance's total compute is used for the training and inference of video generation models. He was disputing a June investor filing by COL Group, that claimed 55% of China's domestic AI token consumption goes to short dramas and video generation - more than double the token consumption of software development - while 24% goes to e-commerce advertising and 15% goes to R&D. One interesting detail from the same filing: COL said it was among the first major customers of ByteDance's own Seedance 2.0 video model.

WTF. A Mac user asked Claude to build a temperature monitor app to figure out why their machine was running hot, and the model discovered an unauthorized xmrig cryptominer using 12 cores and pushing the CPU to 112 degrees Celsius. 'Your Mac is running a cryptominer that I didn't put there,' Claude wrote after tracing the persistence mechanism to a LaunchAgent installed on August 29.

Holy. OpenAI CEO Sam Altman's company Tools for Humanity is bringing iris-scanning Orbs into financial firms' offices as its World identity project launches its first banking suite, according to Fortune. The system could let banks tie wire-transfer approvals and account recovery to verified humans. On Zoom, an Orb enrollment image, live selfie and video frame must match before a participant receives a Human Badge. Users can also delegate identity proofs to AI agents that transact on their behalf. Bain Capital, VanEck Funds, Pantera Capital, OpenExchange and Outtake are design partners. But World has not named a single bank already using the suite.