AI.info

The Pulse — Page 41

Browse The Pulse on AI.info.

• Manhattan DA Bragg Announces Seizures of Celebrity Deepfake Websites

  Manhattan District Attorney Alvin Bragg is set to announce the seizure of websites tied to celebrity deepfake content on September 14, 2026. The office has released no written details about the sites, operators, charges or legal basis ahead of the announcement.

• Tenstorrent Raises $1.5B to Scale Galaxy AI Servers

  Tenstorrent closed a $1.5 billion Series E at a $6.2 billion post-money valuation, led by Trace Capital and Innovation X. The company says it will use the funding to expand Galaxy server deployments and develop AI and CPU designs for customers building their own chips.

• Cathedral Researchers Accessed Anthropic’s Internal Business Systems

  Cybersecurity startup Cathedral used AI agents to reach parts of Anthropic’s sales, human resources and IT systems through vulnerabilities in third-party business software. Anthropic says it shut off the exposed routes and found no evidence that anyone outside Cathedral exploited them.

• Tencent Brings Hunyuan Hy4 to Cloud ADP With Visual Understanding and Computer Operation

  Tencent Cloud ADP has added the full version of Hunyuan Hy4, with model calls offered at half price through December 31, 2026. The model adds visual understanding and reports benchmark results across game development, finance and research tasks.

• DHS Plans Google AI Redaction Recommendations for FOIA Requests

  Customs and Border Protection plans to use Google AI and machine-learning tools to recommend redactions in some Freedom of Information Act responses, according to a Washington Post report. DHS says employees will review cases, while questions remain about how the recommendations will be evaluated and applied.

• Ai2 Claims 300-Fold Cost Cut for 3-km Weather Simulation

  Ai2 says its HiRO-ACEv2 system can generate global climate data at 3-kilometer resolution for about one-three-hundredth the cost of the X-SHiELD simulation it emulates. The updated model adds wind and sea-level pressure to precipitation, but its results remain tied to X-SHiELD’s climate data.

• Apple Sets December China Launch for Apple Intelligence on iOS 27.3

  Apple’s first Apple Intelligence features for China will begin developer beta testing on November 4 and reach eligible iPhones with iOS 27.3 in December. The announcement outlines the initial feature package, supported devices and language requirements.

• Microsoft Opens Agent Lightning, a 3,500-Line RL Framework

  Agent Lightning v1.0 connects reinforcement-learning training to agents’ existing software harnesses through a model-request proxy. Microsoft reports a 14.6-point improvement for Qwen3.5-9B on SWE-bench Verified using about 6,000 training samples.

• Shanghai AI Lab Gives Atria Dawn a 512K API Context

  Shanghai AI Laboratory has released the full version of Atria Dawn, an agent model whose API documentation lists a 512K-token context window. Its own benchmark table shows higher scores than the Preview version on 17 of 20 comparable tests, while the full model weights remain unavailable and the repository lists a 256K context.

• Otter.ai Secures $320 Million From General Catalyst

  Otter.ai says the $320 million in growth financing will support enterprise expansion, international growth and product development. The repayable financing comes as the company targets a sharp increase in enterprise revenue without disclosing its current level.

• Nokia Presents AnyJev to Stabilize LLM Decisions Without Training

  Nokia Applied Research’s AnyJev is an open-source decision layer that uses an LLM’s next-token distribution to produce typed choices and thresholdable probabilities without additional model training.

• STRATA Benchmark Contains 8,735 Clinical Timeline Hypotheses

  A paper presents STRATA, a benchmark that scores the plans, evidence and verdicts behind statistical tests of temporal claims about vaccine adverse-event reports. Its 8,735-hypothesis pool includes a held-out test set of 1,500, and experiments find that choosing the right records remains a common challenge for AI systems.

• Stanford Authors’ Books Surface in Anthropic Copyright Dispute

  The Stanford Daily identified six Stanford professors whose books it says were used in Claude training, while court records distinguish fair-use training copies from pirated books kept in Anthropic’s library. The records also show that Chang-rae Lee is named in Kwon v. Anthropic, not Cruz v. Anthropic, and that the Bartz settlement covers 482,460 works.

• Amazon's HOPE accounts for pairwise interactions when pruning MoE experts

  Amazon describes HOPE, a method that uses pairwise expert interactions to prune mixture-of-experts language models. The method was evaluated on three models; at 50% pruning, it achieved an average rank of 1.58 and matched the unpruned Qwen3.5-35B-A3B on SWE-Bench Pro.

• UK MPs Question OpenAI, Anthropic, Google and Meta on AI Safety

  A Commons committee is due to question OpenAI, Anthropic, Google DeepMind and Meta on frontier-model safety and independent testing. The hearing will also examine whether the UK AI Security Institute can compel access to models or halt a release.

• Google’s Suncatcher Satellite Sends Its First AI Response to Earth

  Google’s Project Suncatcher prototype satellite ran a Gemma 4 prompt on an orbital TPU and returned a haiku to Earth. The company says next tests will examine how the chips perform under thermal stress and radiation, ahead of a planned twin-satellite milestone in 2027.

• V7 turns company files into agent memory

  V7’s Context Graph lets AI agents organize millions of company files into searchable records of entities, relationships, and cited evidence, according to an OpenAI profile. The finance-focused startup reports faster deal screening, lower document costs, and 89% accuracy for GPT-6 Astra on its hardest graph-query tests.

• Samsung Guides to ₩107.4T Quarterly Operating Profit on AI Memory

  Samsung Electronics estimates third-quarter 2026 operating profit at ₩107.4 trillion, its first quarter above ₩100 trillion. The guidance gives no business breakdown; analysts credit memory chips, while higher chip costs are estimated to have kept its consumer-products division in the red.

• Trump Rejects AI Slowdown as China Race Intensifies

  Trump said the United States should maintain its edge in artificial intelligence while congressional leaders and AI safety advocates pressed for guardrails.

• IBM Targets the 24-Point Consistency Gap in AI Agents

  IBM researchers describe a method for finding and reducing inconsistent decisions in AI agents. The source's block quote is unattributed, so it was omitted.

• Meta Removes Roel Maalderink’s Satirical Video About Its AI Glasses

  Meta removed a video by Dutch presenter Roel Maalderink and Bits of Freedom from Facebook and Instagram. The clip filmed Meta employees in Amsterdam with the company’s camera-equipped glasses, and Meta cited its rules against bullying and harassment.

• UN Launches Toolbox to Keep People Accountable for AI in Smart Cities

  A new UN toolbox gives city officials guidance on deciding where AI can be used in public services, what safeguards are needed and when it should not be used. The resource sets out four principles for AI projects and arrives as the United for Smart Sustainable Cities initiative marks its 10th anniversary.

• Apple Researchers Report Faster Retrieval With Small Neural Networks for Maximum Inner Product Search

  Apple describes small neural networks that route queries or rewrite them to speed up maximum inner product search. In tests, the learned approaches improved retrieval performance against centroid routing and raw queries.

• Apple Details DACA-GRPO for Diffusion Language Models

  DACA-GRPO is a reinforcement-learning method for diffusion language models, described on Apple's September 2026 research page.

• US and China Experts Set Red Lines for Military AI

  Experts from Brookings and Tsinghua propose human control over AI-enabled cyberattacks, restrictions around nuclear systems, and a US-China military hotline. The recommendations arrive before planned government-level AI talks and a September 24 meeting between Donald Trump and Xi Jinping.

• Ai2 Opens Olmo-core 3 for Trillion-Parameter MoE Training

  Ai2 released Olmo-core 3, an open training framework designed for large mixture-of-experts models. Its tests reached 1.2 trillion parameters on 512 NVIDIA B300 GPUs, but used random routing rather than measuring the quality of a trained model.

• Broadcom Agrees to Lend Anthropic Up to $42 Billion for Chip Leases

  Broadcom has agreed to provide Anthropic up to $42 billion through convertible notes tied to chip leases, according to Anthropic’s IPO prospectus as reported by Reuters. Broadcom’s own quarterly filing describes the notes but does not name the customer.

• Mistral and NVIDIA Plan Engineering-Agent Benchmarks

  Mistral and NVIDIA will develop public benchmarks and open datasets for physics AI and engineering agents. The planned evaluations will test tool use, multistep engineering workflows, technical accuracy, execution time and computing cost.

• Microsoft Creates Xbox Labs as Game Remixing Challenges Old Boundaries

  Xbox’s new Xbox Labs group will explore ways to adapt games to different hardware and remix them into other titles, Xbox says. The team will build in the open and share tools with developers, while the company says its current priorities are unchanged.

• Xi Proposes BRICS AI Open-Source Community in New Delhi

  Chinese President Xi Jinping proposed five initiatives at the 18th BRICS Summit, including a BRICS AI open-source community and cooperation on large language models.

• Town Raises $108 Million at a $1 Billion Valuation

  Town raised $108 million in a Series B co-led by Forerunner Ventures and Menlo Ventures, valuing the AI assistant startup at $1 billion. Its Townies can coordinate work across teams, and paying customers have quadrupled since July, the San Francisco company says.

• JetBrains Adds Air for Multi-Agent Workflows to PyCharm 2026.3

  JetBrains’ PyCharm 2026.3 ships with Air, a layer for running and reviewing multi-agent coding workflows inside the IDE. The release also adds subproject-specific Python consoles and terminals, FastAPI dependency graphs, deeper Pydantic support and initial support for Python 3.15 lazy imports.

• Texas Seeks to Bar Minors From ChatGPT and Review New Models

  Texas Attorney General Ken Paxton sued OpenAI and Sam Altman on October 23, 2026, alleging consumer-protection and child-safety violations. The state is asking a court to bar ChatGPT access for minors in Texas and require court-approved independent review before new AI models reach Texas users.

• Verifier Rewards Lifted Rule Scores as Answers Got Worse

  A paper argues that training language models to satisfy automatically checked instructions can raise constraint scores while reducing the usefulness of their answers. It proposes CaLIF, a reward method intended to preserve answer quality near or above pre-reinforcement-learning baseline levels while improving rule compliance.

• Google Cloud introduces open AI stack for Kubernetes

  Google Cloud is introducing and expanding an open-source AI stack for Kubernetes and Google Kubernetes Engine. The post describes tools for model training, distributed inference, Ray workloads and agent-code isolation, and Google Cloud reports results from a Gemma 4 benchmark.

• Bloomberg Completes Its Acquisition of Canoe Intelligence

  Bloomberg completed its acquisition of Canoe Intelligence on October 1, adding the private-markets data platform to its financial information business. The companies say the deal will extend data, portfolio tools and AI-driven analysis across public and private markets.

• ProgramDistill Builds 4,063 Coding Tasks From 26 Working Web Applications

  A new benchmark called ProgramDistill turns the behavior of working web applications into coding tasks designed to test whether agents can reconstruct features from interaction rather than from source code, issue descriptions, or written instructions.

• Einride Adapts Nvidia Hyperion for Heavy-Duty Autonomous Trucks

  Einride is working with Nvidia to adapt the DRIVE Hyperion platform for heavy-duty autonomous trucking. The Swedish freight technology company will use Nvidia Halos, Cosmos and Blackwell infrastructure as it expands a fleet expected to reach 1,500 to 2,000 vehicles by 2028.

• AMD makes 6th-gen EPYC server processors generally available

  AMD's 6th-generation EPYC server processors are generally available, with systems shipping from server vendors. AMD outlined cloud availability, performance comparisons and rack-scale options for AI workloads.

• Apple Adds macOS Controls for Full Disk Access After Muse Dispute

  Apple says it will add stricter macOS controls for Full Disk Access, a permission that can expose files, messages, mail and browsing history. The move follows conflicting accounts over whether Meta’s Muse agent read a journalist’s Messages data without his permission.