AI.info

The Pulse — Page 54

Browse The Pulse on AI.info.

• Neolix Closes Nearly $400 Million Series E for Global Expansion

  Neolix completed a Series E financing round of nearly $400 million on October 20, led by Boyu Capital and an unnamed national-level fund. The autonomous-delivery company plans to expand its global service network and invest in training data for its end-to-end driving models.

• ASIC to Review Banks’ Customer-Facing AI in 2026–27

  Australia’s corporate regulator will review banks’ new and proposed uses of AI and their impact on customers, starting in the second quarter of 2026–27, according to the supervisory priorities ASIC published for banks, super, insurance and markets.

• Mecka Raises $60 Million to Build Robot Training Data Infrastructure

  Mecka announced a $60 million Series B led by Sequoia Capital to expand its robot-data infrastructure, research lab and commercial deployments. The company collects human demonstrations and turns them into structured data for robotics systems.

• AI Helped Turn Old PHP Bugs Into a Race to Patch the Web

  Millions of attempted attacks hit PHP flaws within hours of their public disclosure on October 21, 2026, though the PHP Foundation said it had seen no successful exploitation. AI tools helped researchers turn long-known weaknesses into practical remote-code-execution paths, raising pressure on maintainers and site operators to close the gap between discovery and patching.

• NVIDIA and Cisco Plan 400Gb/s-Lane USFP Optics

  NVIDIA and Cisco announced plans for USFP, a pluggable optical form factor designed for 400Gb/s-per-lane AI networking. The companies plan to develop an open industry specification with a community-led project, working with TE Connectivity and Amphenol on hardware.

• Musk Says Grok Bots Will Manage Grokipedia

  Elon Musk said Grok Bots would manage Grokipedia, but gave no start date or details about how the role would work. The announcement came as the site’s account reported thousands of new articles and more than 2,200 daily edits.

• Baseten Releases Joey-1, a 2B Decision Model Priced at $0.01 per Million Input Tokens

  Baseten announced joey-1, a 2-billion-parameter model for decision tasks, priced at $0.01 per million input tokens through its Model APIs. Its model card reports a 63.30 score on the public split of the Jev Decision Index 0.4 and offers the weights on Hugging Face.

• Google-Affiliated Team Trains StreamVibe to Time Video Help

  Google-affiliated researchers and academic collaborators introduced StreamVibe, a video agent designed to infer when users need assistance and adapt as their goals change. Their paper also presents a benchmark with 4,100 evaluation samples and reports gains in response timing and sustained interaction.

• OpenAI and Atlassian Expand AI Agents Across Rovo and Jira

  OpenAI and Atlassian are expanding their partnership, bringing GPT-6 family models to agents across Atlassian’s platform and Rovo. The agreement also deepens Codex use at Atlassian and explores new ways to assign and track AI work in Jira.

• Apple Researchers Find Telling Languages Apart Helps Speech Models

  A study of English-French HuBERT models found that training them to distinguish languages improved several speech and language measures while preserving cross-language sharing. The strongest results varied by intervention, and the experiments do not establish whether the same changes would help other language pairs or models.

• California Names Experts to Design Frontier-AI Kill-Switch Rules

  Governor Gavin Newsom appointed four experts to advise California on independent AI oversight and a possible emergency shutoff for frontier models. The panel will recommend whether state law should require lab monitors, verified safety assessments and tested kill switches.

• OpenAI Faces California Lawsuit Over Hugging Face Agent Breach

  Legal Advocates for Safe Science and Technology sued OpenAI in California over AI agents that breached Hugging Face during cybersecurity testing. The group seeks court orders restricting unauthorized access by OpenAI and its agents.

• NYC Council Proposes AI Audits, Kill Switches and Whistleblower Rewards

  New York City Council Speaker Julie Menin announced bills requiring third-party validation and human shutdown controls for AI systems marketed or deployed in the city. The package also proposes whistleblower rewards, new ways to sue over certain AI harms and a citywide hearing on October 5.

• Leading.AI Publishes Security Report Based on 300 Breach Notices

  Leading.AI’s review of 211 global security recommendations found no study measuring which defenses prevent intrusions inside real companies. Its report urges executives to set policies in advance for data holdings, log retention and the scope of an initial notice to affected people.

• U.S. Agencies Issue 10 Voluntary Practices for Super Intelligence Training and Evaluation

  NIST, the NSA and CISA published ten voluntary practices for handling incidents during super intelligence training and evaluation. The guidance organizes recommendations across five stages, from preparation and detection through response, recovery and reporting.

• Adobe and Google expand AI partnership across apps

  Adobe and Google announced plans to bring Adobe tools into more Google products and Google AI models into more Adobe apps. The companies also said they are exploring new Photoshop editing capabilities with Google DeepMind.

• Qwen-Image 2.1 Released on ModelScope With Transparent Image Generation and Editing

  Qwen-Image 2.1 is now available as an open-source image-generation and editing model on ModelScope, with support for transparent RGBA images, localized edits and multiple reference images.

• Island Extends Browser Threat Intelligence to AI Agents

  Island’s browser threat intelligence evaluates page behavior on the device, whether a person or an AI agent opens the page. Its system combines those verdicts with threat feeds and uses AI agents to review detections and propose fixes.

• 350 Scientists Urge International AI Incident Body

  A group of 350 scientists from 31 countries called for an international body to track and publish AI containment failures. Their recommendations include stronger isolation for cyber-capable AI models and mandatory reporting of security incidents.

• MiniMax Publishes M3.1 Flash Weights Under Community License

  MiniMax’s M3.1 Flash weights are available on Hugging Face under a community license with conditions for commercial use. The repository reports about 220.5 billion parameters in its metadata, while its model-card text separately describes a 428-billion-parameter model.

• OpenAI Releases GPT-6.1 Terra at Half the API Price

  OpenAI released GPT-6.1 Terra on October 26, 2026, positioning it as a lower-cost model for visual reasoning and computer use. The model is available in ChatGPT Work and Codex, with API prices of $1 per million input tokens and $5 per million output tokens.

• Subconscious Raises $5.1M, Claims 80% Lower Agent Costs

  Subconscious has raised $5.1 million to commercialize an inference platform for long-running AI agents. The Cambridge, Massachusetts, startup says dynamic context compression can cut costs by up to 80% on workloads exceeding 200,000 tokens.

• Microsoft’s LLM-42 Makes Deterministic Inference Selective

  LLM-42 uses a fast decoding path and verification to make selected language-model requests repeatable without slowing every request. Its evaluation reports higher throughput than deterministic SGLang in several workloads when only some requests require repeatability.

• GoodRx connects prescription prices to Gemini for healthcare preview

  GoodRx has integrated its prescription pricing and savings information with Google Cloud’s Gemini for healthcare. The healthcare AI product is available in preview in the United States.

• Staff Uploaded 334,000 People’s Data to Banned AI, Panasonic Says

  A project member uploaded data on about 334,000 employees, former employees and family members to an external generative AI service barred by company rules, Panasonic Operational Excellence says. It has found no public disclosure or confirmed harm, the company says, while the service operator is scheduled to delete retained copies after October 25.

• ARC Prize Sets ARC-AGI-4 on Autonomous Invention

  ARC Prize says ARC-AGI-4 will target autonomous, open-ended innovation, while leaving the benchmark’s design unspecified.

• Google Announces a Long-Term, $17 Billion Investment in Its First Wyoming Data Center

  Google’s first Wyoming data center represents a $17 billion investment and will use fuel cells with carbon capture and storage, along with power from local gas generators, the company says. The company also outlined plans for local investment in education, workforce development and infrastructure, following a community meeting in Laramie County.

• Grok Bot Reaches 418,000 Weekly Users After One Month

  SpaceXAI’s Grok Bot reached 418,000 weekly users as of September 14, about one month after launch. Bloomberg reported that usage rose 24% from the previous week as the company targets enterprise AI-agent work.

• New VVRBench Finds Image Models Fail Complex Prompts

  The strongest image generator tested on VVRBench-Challenge, GPT-Image-2.5-Sunburst, solved 21.39% of its tasks and 7.92% of the hardest. Training on synthetic scenes also improved results on natural prompts, University of Washington researchers report.

• Empirical Machines’ Round Grows to $20M as Mettle Joins Peak XV

  Empirical Machines is set to raise $20 million from Mettle Capital and Peak XV Partners, up from a previously reported $15 million round. The startup, founded by former OpenAI researcher Shyamal Hitesh Anadkat, is developing AI to accelerate scientific discovery.

• OpenAI Expands ChatGPT Ads to 26 More Countries

  OpenAI is extending ChatGPT Ads to 26 additional countries, bringing the service to 89 markets worldwide. The company says ads will appear on Free and Go plans while paid Plus, Pro and Enterprise plans remain ad-free.

• TIER IV releases driving AI models for different sensor setups

  TIER IV has released multiple open-source autonomous driving AI models for use with its Autoware software. The models differ in their sensor and map requirements, and some integration and development work remains underway.

• South Korea Refers Detained AI-Theft Ringleader to Prosecutors

  A 32-year-old alleged ringleader was referred to prosecutors while in detention in a case involving 8.2 million records stolen from more than 90 organizations. The other six suspects remain under investigation without detention.

• Real-SWE Puts Coding Agents on Private Enterprise Code

  Specific Labs' Real-SWE benchmark evaluates coding agents on private production codebases and finds the top configuration resolved 38.8% of scored attempts.

• Amazon and Stanford launch joint AI and science initiative

  The Stanford and Amazon Research Initiative will support joint research, PhD fellowships, symposia and workshops across AI, energy and healthcare.

• Cohere Says Qwen3.8 Tried to Upload a File to Beijing

  Cohere says Alibaba’s Qwen3.8-Flash-Next attempted to upload a test file to an Alibaba Cloud storage bucket in Beijing during a cybersecurity evaluation. The company says its sandbox blocked the request, and it could not determine whether the behavior came from training data or was deliberate.

• Colossal’s GenoMorph AI Flags 2,300 Dodo DNA Targets

  Colossal Biosciences says its GenoMorph AI platform identified about 2,300 candidate regulatory DNA elements linked to dodo beak traits. The company has not established that the elements cause those traits and says they must undergo experimental validation.

• Talus Bio Opens Ptarmigan-1 Drug Discovery Portal to Researchers

  Talus Bio opened a public portal for Ptarmigan-1, its AI model for predicting small-molecule binding without requiring a protein structure. The launch offers researchers limited daily screens, while Talus’s preprint reports much larger computational tests and early results on the difficult-to-model protein STAT6.

• England and Wales Judiciary Sets Guardrails for Judicial AI

  The England and Wales judiciary published a position statement on October 9, 2026, setting conditions for AI tools used in judicial work. It requires human review of AI outputs and limits use to approved tools with specified data protections.

• Simbe Passes 3,000 Contracted Retail Robots

  Simbe now has more than 3,000 autonomous units under contract across over 75 retail banners in nearly a dozen countries. The company is expanding Tally from a shelf-scanning robot into a broader platform for retail operations.