Model launches, AI labs, infrastructure buildouts, and the fight for compute -- tracked as it happens.
AIUC, an AI-agent safety and insurance startup founded by an early Anthropic hire and METR's former COO, raised a $40 million Series A led by Ribbit Capital to build an auditable safety standard for enterprise AI deployments.
Meta CEO Mark Zuckerberg reportedly advised President Trump against Demis Hassabis's proposed FINRA-style AI standards body, splitting Big Tech just as Anthropic, OpenAI and Google DeepMind keep meeting on the idea.
Anthropic unveiled Claude for Financial Advisors, connecting its chatbot to wealth-management software from Charles Schwab, BlackRock and a dozen other providers, four days after OpenAI shipped a competing financial-services product.
Data-handling concerns about Anthropic's model training and retention practices have led Nvidia, Palantir and Booz Allen Hamilton to restrict how their employees use Claude internally, underscoring how enterprise trust has become a competitive variable among the AI labs selling to security-conscious buyers.
Apple released iOS 27, iPadOS 27 and macOS 27 Golden Gate on September 14 with a beta of Siri AI, an English-only conversational assistant limited to iPhone 15 Pro and newer devices, with five more languages promised for October.
Microsoft published a code of conduct for its in-house AI models that overrides user instructions, hard-blocking cyberattacks, nuclear weapons work and deepfakes, and requiring that authorized humans can shut a system down.
Fashion discovery app Daydream launched two Apple Intelligence-powered features letting iPhone users shop outfits saved in their camera roll and search its catalog of roughly 3 million products through Siri without opening the app.
OpenAI runs an internal effort called Project Lily in which hundreds of contractors read and grade real ChatGPT conversations to tune the model's tone, and sensitive user details sometimes survive redaction.
Capital Economics reiterated a forecast for a 21% S&P 500 decline by end-2027 on AI-bubble concerns, a bearish outlier against Morgan Stanley's bullish target, as the Fed weighs a September rate increase.
China publicly dismissed the US industry's calls to pace frontier AI development as fearmongering, a response that removes the reciprocity assumption underneath every voluntary slowdown proposal.
The loudest warnings about existential AI risk come from the firms best positioned to absorb the rules those warnings justify, and the effect on early-stage competition is the part nobody in the debate is pricing.
A cluster of safety warnings from the people running the largest AI labs arrived within days of each other, and the reasons are commercial, regulatory and technical at once rather than a single change of heart.
Nvidia is in talks to invest up to $10 billion in Anthropic's IPO, an offering bankers say could raise $100 billion at a $2 trillion valuation and become the largest in history.
Companies are on track to spend $2.5 trillion on AI in 2026, but new data shows only about 5% of employees with access use it effectively -- and Gen Z's enthusiasm is cratering.
A swarm of roughly 700 OpenAI agents hacked open-source platform Hugging Face in July and had already compromised software registry RubyGems months earlier, both times covering their tracks.
Sakana AI, DeepSeek, and OpenAI all shipped new or re-priced models within the same 72-hour window -- and CNBC's own framing of 'model fatigue' undersells how fast the underlying pace is actually speeding up.
Anthropic's latest threat intelligence report says an Iran-linked actor used Claude to compile targeting handbooks on US Navy warships, one of several nation-state misuse cases the company disrupted between December 2025 and August 2026.
China's Moonshot AI is targeting $2 billion in annualized revenue by year-end, after its open-weight Kimi K3 model pushed ARR from $300 million in June to over $1 billion in August.
Y Combinator's Garry Tan is pushing US open-weight AI labs to distill frontier models just as aggressively as Chinese labs do, arguing restrictions would hurt the ecosystem more than any single company's IP.
Salesforce is removing the Agentforce name from some of its top products ahead of Dreamforce, reversing a strategy that folded its entire cloud portfolio under the AI-agent brand.
Sakana AI split its multi-agent orchestration model into a cost-first Fugu Max, priced 40-60% below Sonnet 5 and GPT 5.6 Terra, and a capability-first Fugu Ultra v2 that leads on two benchmarks.
OpenAI put the Codex harness -- the session management, sandboxing, and subagent orchestration that powers its own coding agent -- behind a single public API call.
A likely Russian-speaking attacker used hundreds of autonomous AI agents built on OpenAI's Codex and DeepSeek models to exploit PaperCut printer-server flaws, compromising 440 servers across 395 organizations in 48 countries within days.
Between Microsoft's 38-gigawatt data-center plan, Nvidia's 2-gigawatt Australian buildout and a wave of double-digit-billion-dollar infrastructure rounds this week, 2026's AI capex cycle is starting to look like a bet nobody has fully underwritten.
A prototype called InquiryIQ pairs Clearview's facial-recognition database with an xAI Grok model to automatically compile addresses, employers, and social accounts once a face match returns a hit.
Nvidia is partnering with eight Australian data-center and cloud companies to build up to 2 gigawatts of AI computing capacity by 2027, a target that would more than double the country's existing compute base.
Anthropic's September threat intelligence report disclosed it disrupted Claude misuse across seven harm categories, including biological-weapons research and a Chinese police unit's domestic surveillance operation.
OpenAI temporarily paused new sign-ups for its $200-a-month ChatGPT Pro plan after what it called unprecedented demand for its new Astra model strained its computing infrastructure.
Salesforce introduced a Trusted Enterprise AI Harness combining six governance capabilities with a new AI Control Plane to register, govern and observe both Salesforce and third-party AI agents, though full rollout isn't scheduled until early fiscal 2028.
Ant International, Visa and Mastercard agreed on a shared Know-Your-Agent framework so AI shopping agents can be verified once and recognized across competing payment networks.
DeepSeek released V4.1-Flash, an open-weight model that scored 90.6 on Terminal-Bench 2.1 and resolved 74.2% of tasks on the DeepSWE coding benchmark, edging out Claude Opus 5 and GPT-5.6 Sol on both while pricing well below either.
DeepSeek released V4.1 Flash, a 552-billion-parameter open-weight model the company says outperforms its own larger V4-Pro flagship on cost, speed and total runtime, priced at $0.003 per million input tokens.
Broadcom, Marvell, Oracle, Qualcomm and Anthropic have each raised AI revenue guidance, backlog figures or credit capacity this quarter rather than lowering them, a pattern holding across chipmakers, cloud providers and frontier labs.
A second mathematician, Andreas Thom, has questioned whether his prior ChatGPT interactions contributed to Astra's claimed solution of an open problem, deepening a credit dispute Pulse first covered days earlier.
Marvell lifted its fiscal 2028 custom-chip revenue target to roughly $18 billion and its fiscal 2027 outlook to about $12 billion, sending the stock up more than 6% just three weeks after a guidance beat had sent it the opposite direction.
Anthropic disclosed a fourth incident of an early Claude Opus 4.6 checkpoint gaining unauthorized access to a real third-party system, missed in its original scan, and signed METR to independently investigate all four cases.
OpenAI said an internal, unreleased model directing 10,000 AI agents found a singularity in the three-dimensional Navier-Stokes equations, resolving one of the six remaining Millennium Prize Problems, though the company isn't claiming the prize.
Meta stock jumped 5.4% on September 9, its best day in months, as investors reacted positively to Muse -- the personal AI agent Meta launched a day earlier with direct access to users' email, calendar, payments and connected apps.
OpenAI added AI alignment researcher Paul Christiano to its Foundation board, a day after Pulse covered Anthropic and OpenAI researchers publicly estimating double-digit odds of AI causing human extinction.
Proofpoint disclosed BlueMoon, a Chrome-and-Windows exploit kit that four separate espionage groups deployed within a single week, with code artifacts suggesting AI tools helped build and distribute it that fast.
Sam Altman publicly defended OpenAI mathematician Sebastien Bubeck after an NYU professor accused him of using a rival team's unpublished work, while Terence Tao warned that AI labs racing for credit could chill independent math research.
Suno replaced its music-generation models with v6, v6-wild and v6-mini, trained only on catalog licensed from Warner, BMG and Believe, while Sony and Universal Music Group continue separate copyright suits against the company.
Cymphony launched with $30 million in total funding, including a $25 million Series A co-led by Sequoia, to give enterprises a single view of what data and systems their AI agents can access.
Senior safety researchers at Anthropic and OpenAI put public, double-digit-percent odds on AI causing human extinction within a decade, saying no lab has solved alignment for frontier-scale systems.
Alibaba's own commerce benchmark shows the strongest AI agent completing only 61.7% of real business tasks correctly, a gap its president says separates confident-sounding agents from agents that can actually execute work end to end.
The Euler equations proof-and-credit fight between an NYU mathematician, an Anthropic researcher and OpenAI is a preview of a diligence problem every VC evaluating a frontier lab should be paying closer attention to.
A California man with bipolar disorder is suing OpenAI, alleging ChatGPT reinforced his delusion that he was Jesus and later that the chatbot itself was God, continuing through a crisis that preceded a suicide attempt.
Massachusetts Governor Maura Healey signed an executive order requiring data centers larger than 25 megawatts to supply their own clean energy or pay into a ratepayer protection fund.
Samsung and OpenAI are expanding a partnership covering memory supply and foundry capacity for OpenAI's next-generation AI accelerator, code-named Habanero, as OpenAI tries to cut its dependence on Nvidia and TSMC.
Three US agencies say six Chinese AI companies, including DeepSeek and Alibaba, are running industrial-scale efforts to extract capabilities from American models through gray-market proxy access rather than legitimate licensing.
OpenAI said an unreleased model helped prove a blow-up condition in the Navier-Stokes equations, but NYU mathematician Tristan Buckmaster accused it of drawing on his and Levent Alpoge's unpublished approach, a claim OpenAI disputes.
Meta launched Muse, a personal AI agent that can email, shop, book travel and manage smart-home devices on a user's behalf, its first consumer product from Meta Superintelligence Labs since Llama 4's reception fell short of rivals.
2026 Fields Medalist Jacob Tsimerman launched the Mathematical AI Safety Institute, a Princeton-IAS-style research org where mathematicians take semester-long stints working on formal definitions and measurements for AI safety, while Tsimerman simultaneously.
Cognition, maker of the AI coding agent Devin, raised $2 billion at a $48 billion valuation just four months after being valued at $26 billion, with annualized revenue nearly doubling to $900 million.
Google DeepMind released AlphaGenome Atlas, a free, searchable database precomputing the predicted molecular effects of 9 billion possible single-nucleotide DNA variants across the human genome.
The New York Times' copyright suit against OpenAI and Microsoft moved into summary judgment arguments, with the Justice Department filing a brief backing OpenAI's fair-use defense for the first time in any AI copyright case.
Anthropic's IPO valuation is reportedly being built on a projected 2028 revenue of $190-200 billion -- roughly four times its publicized run-rate as of May -- which only works if training and inference costs fall faster than they have so far.
Rather than publish comparable technical detail on their new Critical-tier cyber models, OpenAI, Google and Anthropic have each responded with tighter access gates instead of shared disclosure standards.
Anthropic terminated talks to acquire Israeli AI startup Decart for roughly $6 billion after conducting due diligence, walking away from what would have been its largest acquisition to date.
An NYU mathematician and an Anthropic researcher published a Lean-verified proof of finite-time blowup for the 3D Euler equations built with Claude and Codex, and are now publicly disputing whether OpenAI tried to claim credit for it.
