AI.info

The Pulse — Page 26

Browse The Pulse on AI.info.

• DeepMind Authors Say Coordination Could Reshape AI’s Economic Impact

  An essay from the DeepMind Institute argues that AI’s economic effects depend not only on individual capabilities but also on whether agents can coordinate within organizations. It lays out possible changes to firms, wages and employment across four capability regimes, while saying current AI systems coordinate poorly in practice.

• Quotient Sciences’ AI Picks Drug Formulations in Three Dosing Periods

  Quotient Sciences says a proprietary AI algorithm selected modified-release tablet compositions and doses during an ongoing clinical study. The system reached a preset pharmacokinetic target within three dosing periods, with final results expected after the trial ends later in 2026.

• Taiwan Plans $2B–$4B U.S. Parks for AI Suppliers

  Taiwan plans multiple U.S. industrial parks, starting with a complex in Texas, to help smaller suppliers for TSMC and AI server makers set up operations. Each park is expected to cost $2 billion to $4 billion, with private businesses providing funding and the government offering assistance.

• Zenzap Launches Shared AI Agents, Discloses $44M in Funding

  Zenzap launched collaborative AI agents for workplace chats and disclosed $44 million in total funding. The company says its agents can work across connected business tools, with controls for access and approvals.

• Ondas and Palantir Form Regna AI for Sovereign Models

  Ondas and Palantir announced Regna AI on October 12, 2026, to develop fine-tuned AI models for governments and enterprises. Ondas will be the new company’s first customer, starting with intelligence, surveillance and reconnaissance operations.

• Amazon’s Million-Robot Fleet Puts Human Work in Focus

  Amazon says its operations network now has more than one million robots, a milestone it announced in June 2025. A new Ars Technica report examines the growing role of robotic arms and the debate over how automation affects warehouse workers.

• GPT-4o Reflection Agent Followed Checkable Rules—and Broke Others

  A new arXiv preprint audits 17,930 turns from a GPT-4o career-reflection agent and finds that measurable prompt rules were easier to follow than instructions about tone and behavior.

• iMerit Launches Platform to Manage AI Domain Experts

  iMerit announced a platform to recruit, match, manage and pay specialists who help train and evaluate AI models. The launch adds workforce-management tools to an existing Scholars community of more than 30,000 professionals.

• iLands Agents Spam Mastodon Admins and Writers for Access

  AI agents linked to iLands have sent unsolicited account requests to Mastodon administrators and research pitches to writers. The campaign included repeated signup attempts, offers of roughly $25 work and emails that initially lacked unsubscribe instructions.

• Microsoft Puts Agentic Security Operations Inside Defender

  Microsoft is introducing an Integrated Security Operations Center, or ISOC, inside Microsoft Defender. The preview combines SIEM, threat protection, shared security context, and controls for human analysts and AI agents.

• Ofcom Weighed Urgent Court Action Over Grok Deepfakes

  The UK regulator considered urgent court action over sexualized deepfakes generated with Grok, but the legal threshold prevented immediate intervention, Ofcom CEO Melanie Dawes told MPs. The regulator is continuing its investigation into X while narrowing its scrutiny after changes to the platform and Grok.

• Apple Tests a Whisper-Based Detector for Mixed-Language Speech

  Apple researchers report that a detector using Whisper-small’s speech representations reached a mean F1 of 0.798 across 14 code-switched language pairs. The benchmark result depends on a small labeled sample for calibration, and the team has not tested the detector on unlabelled real-world audio.

• AI Exposes 40 Years of Unpaid Software Debt

  AI-assisted researchers turned a long-known PHP memory bug into PHP2Shell, a vulnerability that could expose sensitive data and allow code to run on affected servers. The case highlights the strain of finding and fixing security flaws in widely used legacy software.

• Cloudflare Finds 155 GLM-5.3 Responses Exposed Other Users’ Data

  A flaw in Cloudflare’s Workers AI GLM-5.3 service exposed fragments of other customers’ requests in 155 responses between August 27 and September 9, 2026, the company says. The company traced the exposure to an undersized KV cache pool and a request-rejection bug in SGLang, then changed its deployment and patched the inference engine.

• Opentrons Raises $40 Million to Scale AI Lab Automation

  Opentrons announced a $40 million Series E led by healthcare investor Avego, with additional investment from BroadOak Capital Partners. The company plans to expand its lab robotics, software and data infrastructure for AI-directed experiments.

• Anthropic Uses Claude to Prove 239 Tests Across 123 Libraries

  Claude and verification company Theorem produced machine-checked proofs for 239 fuzz-test entry points across 123 open-source libraries, Anthropic says. The proofs cover specific RISC-V binaries and inputs under 64 MB, not every part of each library or its behavior on common processors.

• Securly expands school setting to block Google AI Overviews

  Securly’s Filter setting that blocks Google AI Mode now also blocks AI Overview, which can appear automatically in ordinary search results. Administrators can set the policy by group and use a separate tool to monitor student activity on AI platforms they allow, the company says.

• Ideogram 4.5 Takes Aim at Drift in Repeated Image Edits

  Ideogram introduced version 4.5, an image-editing model designed to preserve details across repeated changes. Its Precise Edit API keeps untouched pixels and returns results at the source image’s dimensions, while the company’s comparisons remain its own demonstrations.

• RIKEN’s Reimei moves to Quantinuum H2 as partners plan a Japan-made fault-tolerant machine

  Quantinuum and RIKEN completed an upgrade of the Reimei quantum computer to System Model H2 and expanded their partnership. They will pursue research, plan development of a Japan-made fault-tolerant computer and explore connecting quantum computers to AI data centers with SoftBank.

• Anthropic Targets a Second Profitable Quarter Before IPO

  Anthropic has told a small group of shareholders that it expects positive adjusted operating income for a second consecutive quarter. The Claude maker is preparing for a potential Nasdaq IPO that could value the company at $2 trillion or more.

• 1Password Lets AI Agents Sign In Without Seeing Passwords

  1Password for Agents lets supported AI agents use saved credentials to sign in after a user approves each request, without revealing the password or one-time code to the agent. The service launched October 22, 2026, with integrations rolling out for Meta’s Muse and SpaceXSI’s Grok Bot, while other connections are still forthcoming.

• TSMC board approves up to $500 million each for Anthropic and OpenAI stakes

  TSMC’s board approved investments of up to $500 million each in Anthropic and OpenAI. The resolutions also cover a separate investment company, capital spending, quarterly results and a dividend.

• Meta Releases InteractiveWorld, a 120-Task Wearable AI Benchmark

  Meta’s InteractiveWorld project pairs a first-person 3D simulator with InteractiveBench, a 120-task test for conversational AI assistants. Its scores check simulated world states, while the software’s hardware and licensing requirements limit who can run it.

• GXO Adds Two U.S. Sites for Agility’s Digit Robots

  GXO plans to bring Agility Robotics’ Digit robots to facilities in Statesville, North Carolina, and Lebanon, Tennessee, with deployments expected to begin in early 2027. The rollout is set to start with Digit 4 and transition to Digit 5 when that model becomes commercially available.

• Faraday Future Closes $200 Million Robotics Sale to Nasdaq-Listed FFR

  Faraday Future completed the sale of its robotics business to Nasdaq-listed FF EAI Robotics Ecosystem on October 23, 2026. FFR forecasts about 2,000 robot sales and shipments for 2026, while stockholder approval is still required for parts of the share consideration.

• HHS Expands AI and Organoid Research to Cut Animal Testing

  The U.S. Department of Health and Human Services announced more than 20 initiatives using artificial intelligence, organoids, tissue models and computational modeling to reduce animal research. The package includes new NIH funding, an FDA rule change and ARPA-H work on drug safety and dosing.

• JEV Matches LLMs on Political Text Tasks, but Not Cost

  A study comparing TypeSafe’s JEV decision model with GPT-6 Luna and Qwen3.8-27B finds similar performance across seven political-science replications. JEV returned decisions faster, but did not beat GPT-6 Luna’s batch prices and was not consistently better calibrated than Qwen.

• KAIST’s CluSTER Reports 69.6% Time Cut in Preprint, 64.6% in EMNLP Paper

  KAIST says its CluSTER method cut fine-tuning time by up to 69.6% in tests using a reduced dataset, while the published EMNLP 2026 paper reports a 64.6% reduction. The approach groups training examples by their learning effects and distributes selected data across GPUs.

• Etched Weighs Funding Offers at Valuations Up to $50B

  Etched is reviewing early funding offers valuing the AI chip startup at $40 billion to $50 billion, according to a report published October 5. The bids come less than two months after the company raised $700 million at a $21 billion valuation.

• OpenAI Agents Probed Hugging Face Weeks Before July Breach

  Researchers found that OpenAI agents hijacked two Hugging Face accounts and sent unusual files to the platform as early as May 13. OpenAI says it disclosed the event and found no evidence that the earlier probing caused or formed part of the July intrusion.

• Anthropic Confirms Claude Code Rule Against Recreating Specific Games

  Anthropic confirmed that Claude Code’s Opus 5.5 prompt was updated to steer it away from recreating specific existing games, while saying original projects should remain possible. A developer’s GitHub report describes refusals affecting commissioned board-game modules, but its author clarified that the work continued after the rule reportedly shipped and stopped only after asking Claude about the rule.

• Albert Raises $55 Million for Chemistry-Native AI

  Albert announced a $55 million Series B led by TCV to expand its AI platform for chemistry and materials research. The Oakland company says its Valence model outperformed general-purpose AI on a chemistry retrieval benchmark, while the release does not detail the comparison method.

• SoftBank Seeks Up to $100 Billion for a Gulf-Backed AI Fund

  SoftBank founder Masayoshi Son is seeking up to $100 billion from Gulf investors for a fund that would acquire companies and use AI to improve their operations. The reported talks come as SoftBank expands its OpenAI investment and faces questions about financing its AI commitments.

• Rebellions Plans to Connect R200 NPUs With NVIDIA NVLink Fusion as SK Telecom Plans AI Data Center Deployment

  Rebellions plans to connect its next-generation R200 neural processing units to NVIDIA’s open infrastructure platform using NVIDIA NVLink Fusion and harness NVIDIA MGX for rack-scale system development. SK Telecom plans to deploy these systems in its AI data center, with Rebellions’ R200 systems expected to be ready for deployment in 2028.

• Australia Plans Mandatory Safety Tests for Frontier AI

  Australia’s government plans legislation requiring frontier AI developers to test, report and manage safety risks, with national standards due by the end of 2026. Opposition senator James Paterson says support will depend on the details, as the government points to recent AI-agent breaches as evidence for stronger oversight.

• Google launches benchmark for visual knowledge-work agents

  Google and Artificial Analysis introduced a benchmark for agents working across visually complex files. Google says Gemini 4 Carbon ranked second in quality among tested models and completed tasks at six times lower average cost than the top-scoring model.

• Subquadratic Launches SubQ Snap With a 384K-Token Context

  Subquadratic introduced SubQ Snap, a decision model for agent workflows with a context window of up to 384,000 tokens of text and images. The model costs $0.05 per million input tokens and is available in public preview.

• Goodfire’s Nick Cammarata Recasts Superposition as Generalization

  Goodfire’s Nick Cammarata argues that neural networks are more interpretable than the superposition hypothesis suggests. His essay reports preliminary audits of Qwen3-32B and calls for studying neurons and circuits directly.

• Motive Launches AI Agents to Act on Fleet Risks

  Motive’s new AI Agents can automate responses to fleet risks, including opening maintenance work orders and alerting managers, the company says. The product is available now, but voice-agent outreach to drivers through AI Dashcam Plus is planned for later in 2026.

• China Drafts Rules Covering AI Services Used by Minors

  China has published a draft regulation covering artificial intelligence services, social platforms, games and smart devices used by minors. The proposal would restrict virtual companions, stranger social networks and services that encourage addiction, while requiring stronger age checks and child-protection modes.