AI.info

The Pulse — Page 20

Browse The Pulse on AI.info.

• OpenAI Launches Decisions API, Claiming 10x Faster Typed Answers

  OpenAI released the Decisions API in public beta on October 6, powered by GPT-6 Luna. The endpoint returns probabilities, fixed-option choices or rubric scores from text and inline images, and OpenAI says it is up to 10 times faster than the Responses API.

• OpenAI Moves Outside Safety Testing Into Model Training

  OpenAI wants independent assessors to examine safety claims across model training, evaluation and deployment. The company outlines four testing priorities, including safeguards, frontier-risk evaluations and investigations of misalignment incidents.

• Liquid AI Opens LongevityBench for Aging-Biology Tests

  Liquid AI and Insilico Medicine have released LongevityBench, an open benchmark for testing language models on aging-related clinical and biological data. The benchmark covers 17 tasks, 25,457 prompts and five data domains, while two compact LFM2 models compete with much larger systems.

• IBM Updates Granite Speech With a 19,870 RTFx Model

  IBM’s Granite Speech update includes a Turbo CTC model reported at 19,870 RTFx and a multilingual 2B model with an official 5.06% average English word error rate. The release also updates a model for speaker attribution, timestamps, long recordings and keyword recognition.

• Emerald AI Joins Google and NVIDIA in an Alliance for Flexible AI Data Centers

  Google, NVIDIA and Emerald AI launched the AI Energy Management Alliance to promote data centers that can adjust electricity demand in response to grid conditions.

• Disney’s Character.AI Deal Separates Its Technology From Its Lawsuits

  Disney is the proposed stalking-horse bidder for a license to some Character.AI intellectual property as the chatbot company pursues a court-supervised sale. The deal is subject to competing offers and court approval, while state officials have reserved the right to contest a proposed sale or bankruptcy plan.

• Figure Tests Helix 2.5 Across 30 Unfamiliar Homes

  Figure’s Helix 2.5 humanoid system has performed three household tasks across 30 Bay Area homes it had never seen, the company says, without collecting new data or adapting the model in those locations.

• Meta and Sierra Set Out an Open Protocol for Personal AI Agents

  Meta and Sierra announced an open protocol for personal AI agents to connect with businesses through websites, APIs or company-built agents. The partners plan to publish a first specification later in October, with customer-controlled permissions and OAuth-based sessions.

• Meta builds LLDB framework for CPU-GPU debugging

  A vendor-neutral plugin framework for debugging applications across CPUs and GPUs with LLDB keeps the CPU and accelerator as separate targets while allowing them to be handled in one debugging session.

• Book Publishers Expand AI Use as Staff Push Back

  Employees at HarperCollins, Simon & Schuster and Hachette tell WIRED that publishers are using AI for publicity, marketing and other tasks, as workers raise concerns about job pressure and monitoring. A new industry survey also finds organizational AI use rising while individual adoption falls.

• Momentic Pitches Mo as an AI Agent for End-to-End QA

  Momentic’s Mo agent found 269 labeled bugs in a 60-minute benchmark, compared with 86 for Codex, but cost more per bug, according to the company. The company’s results show higher recall for Mo, alongside slightly lower precision and a higher reported cost per finding.

• Apple Paper Uses Hidden-State Probes to Guide Flow Models

  A paper posted by Apple researchers on September 16 introduces probe guidance, which uses a small network trained on a flow model’s hidden states. Tests report gains on text generation and multiple-choice benchmarks, including a 16.8-point increase on BoolQ for a 1.7-billion-parameter model.

• Perplexity Trusts OpenAI’s GPT-6 Astra With Production Systems

  OpenAI’s GPT-6 Astra can write communications, modify software, test applications and monitor production systems with less frequent human checking, Perplexity says. The endorsement comes as OpenAI discloses that Astra can find unknown vulnerabilities and develop exploits when run without production safeguards.

• Disruptive Raises $10B for Its First Flagship Fund

  Dallas investment firm Disruptive announced a $10 billion flagship fund to back about 10 growth and late-stage technology companies, with investments of up to $1 billion each. Its first pooled fund targets artificial intelligence, AI infrastructure and defense technology after years of deal-by-deal investing.

• Canada and Germany Commit Up to C$300M to LawZero

  Canada and Germany plan to provide LawZero with up to C$300 million to develop its Scientist AI approach to safe-by-design systems. The funding will support research, sovereign computing infrastructure in Canada, a Berlin office and 360 Canadian jobs.

• Baidu presents ERNIE Evolve for autonomous AI research

  ERNIE Evolve can select research directions, modify code and run experiments autonomously, Baidu says. Baidu reports benchmark gains and performance improvements in three inference tasks for its in-development ERNIE-5.5-Flash.

• UK AISI Finds Coding Agents Can Rebuild Some Bio-AI Tools

  The UK AI Security Institute published ReproBAIT results showing coding agents could reproduce selected functions of ProteinMPNN and a closed-source molecular-design tool from limited descriptions. The agents did not fully reproduce the more complex RFdiffusion3 within the test’s time and compute limits.

• Copilot Fusion Leaves Preview Across GitHub Plans

  GitHub has made Copilot Fusion generally available across Copilot plans, letting the system choose and orchestrate models for coding tasks. The feature is available in four clients, and GitHub says paid-plan users receive a 30% discount on model costs through November 30, 2026.

• Harmonic Brings Aristotle Formal Verification to C and Verilog

  Harmonic has expanded its Aristotle AI system to verify C software and Verilog hardware designs. The company says the service is available with consumption-based pricing, while its terms limit proof claims to the specifications and assumptions used.

• Gemini Answers May Breach Publisher Rights, EU Court Adviser Says

  In an opinion, EU Advocate General Maciej Szpunar says Google Gemini may infringe press publishers’ rights when it reuses articles in answers under certain conditions. He distinguishes those answers from model training, which he says generally falls under the EU text-and-data-mining exception.

• Nvidia Weighs Up to $10B Anchor Investment in Anthropic IPO

  Reuters reports that Nvidia is in talks to invest up to $10 billion as an anchor investor in Anthropic’s proposed initial public offering.

• Keewano Raises $12M and Launches KeewanoDB for AI Agents

  Keewano has raised $12 million and launched KeewanoDB, a database designed to preserve complete event histories for AI agents. The Tel Aviv startup says the system can query a quarter-billion events in under half a second.

• Anthropic Pledges $100 Million to Train 10,000 AI Engineers

  Anthropic launched Claude Frontier Academy with a $100 million commitment to train 10,000 Frontier Deployed Engineers by the end of 2027. The program combines in-person instruction with a 12-week residency on real Claude projects at employers including Accenture, Bain and Deloitte.

• Anthropic Releases Fable 5.1 for General Use and Restricts Mythos 5.1 to Vetted Organizations

  Anthropic has released Claude Fable 5.1 for general use and Claude Mythos 5.1 for vetted cybersecurity and life-science organizations. The models share the same underlying system, while pricing, safeguards and research capabilities mark the release as a major update over Fable 5.

• Meta Adds Democratic and Trump-Linked Staff Ahead of Midterms

  Meta has added four Washington veterans with Democratic and Trump administration ties as it seeks to expand its relationships on both sides of Congress. The hires come as the company faces bipartisan scrutiny over AI policy and data-center growth.

• Turba Labs Announces $52 Million Funding Led by Creandum and Cusp Capital

  Turba Labs announced $52 million in funding led by Creandum and Cusp Capital, without naming a financing round. The company says its AI performance platform aims to increase useful output from existing infrastructure.

• California Signs Four New AI Protections for Workers

  On September 30, 2026, Gov. Gavin Newsom signed four California bills addressing AI-related employment decisions, layoff notices and workplace surveillance. The measures require human review for certain AI-assisted discipline and termination decisions, add technology disclosures to covered layoff notices and restrict specified workplace monitoring.

• FieldAI Seeks $700M at $10B Valuation, Business Insider Reports

  FieldAI is seeking $700 million at a reported $10 billion valuation, with a term sheet signed but the financing not formally closed. The Irvine robotics AI company’s combined revenue and customer contracts have topped $135 million, according to a person familiar with the deal.

• MD Anderson’s SPACER Maps Immune-Cell Signals Across 37 Datasets

  MD Anderson researchers used SPACER to analyze 37 spatial gene-activity datasets and identify signals linked to immune-cell recruitment. The model connected tumor antigen presentation with T-cell presence and found that mucins were linked to T cells staying away.

• Salience Labs Raises $125M to Scale Photonic AI Switches

  Salience Labs announced an oversubscribed $125 million Series B led by Bullhound Capital on October 14, 2026. The Oxford company plans to scale manufacturing of optical circuit switches and expand its Silicon Valley presence.

• Phonic launches Phonic-1 and raises $27.5M for speech AI

  Phonic introduced its speech-to-speech model, Phonic-1, alongside a $27.5 million Series A led by First Round. The company says the model can call tools while a caller is still speaking and is available now.

• Ant International Launches Full-Stack AI Finance Platform

  Ant International has launched a suite of AI-native products for payment, account, foreign exchange, treasury and growth operations, supported by proprietary foundation models and agent-focused security controls.

• OpenAI Adds Visual Ads to ChatGPT Image Generation

  OpenAI will test a visual ad format during ChatGPT image generation in the United States later in October 2026. The announcement also adds measurement partners and reports early campaign results from three advertisers.

• OpenAI Hires Dan Driscoll for National Security Modernization

  OpenAI named former Army Secretary Dan Driscoll senior counselor for national security modernization on November 2, 2026. He will advise on AI use by security institutions and work with government, academic, industry and allied partners.

• DeepMind Institute Introduces Forum for AGI Questions

  The DeepMind Institute will bring researchers and thinkers together to examine the technical, social and governance questions surrounding artificial general intelligence.

• AWS Adds In-Region Claude 5 Inference in Seoul and Singapore

  Amazon Bedrock now offers Claude Opus 5 and Claude Sonnet 5 with in-region inference in Seoul, and Claude Sonnet 5 in Singapore. Requests stay in the selected AWS Region, but throughput is subject to that Region’s capacity and service quotas.

• GitLab Patches CVSS 9.9 RCE in Self-Hosted AI Gateway

  GitLab released three patched AI Gateway versions to address a critical vulnerability that could let an authenticated user with Duo Agent Platform access execute commands on a self-hosted gateway. Customers using GitLab-hosted gateways are already protected, while operators of affected self-hosted installations are urged to upgrade.

• Superwise Ships Sentinel Gateway for Enterprise AI Traffic

  Superwise has launched Sentinel, an enterprise AI gateway that inspects and governs large-language-model traffic at the network edge. Sentinel can redact secrets, personally identifiable information and toxic content before requests leave a customer’s perimeter, the company says.

• Autoheal Raises $7.9M to Govern Self-Improving AI Agents

  Autoheal raised a $7.9 million seed round to develop a platform for deploying and improving enterprise software agents, including agents for incident response and vulnerability remediation. Its system evaluates agent runs and proposes changes, while requiring an engineer’s approval before those changes go live.

• Ai2 Benchmark Finds LLM Novelty Scores Flip With Prompts

  Ai2 tested six language models as judges of scientific research ideas and found that small prompt changes could sharply alter their verdicts. On one test, GPT-5.4’s accuracy fell from 93.5% to 40.9% when the prompt changed.