AI.info

The Pulse — Page 36

Browse The Pulse on AI.info.

• GitHub Makes AI Scan Available for Pull Request Security

  GitHub made AI Scan for pull requests generally available on October 15, adding vulnerability checks for languages and frameworks outside CodeQL’s current coverage. The findings are advisory, and the feature uses Code Security AI Credits.

• Anthropic Pitches Lawmakers a Draft Bill to Restrict AI Shutdowns

  Anthropic representatives pitched lawmakers a draft bill that would limit when the federal government could block the release or use of frontier AI models. The reported proposal includes independent evaluations and would make a court order central to any suspension.

• Anthropic Tests 211 Claude-Designed Scorpion Toxin Binders

  By Anthropic’s account, 17 of 211 Claude-designed protein candidates met its laboratory criteria for binding scorpion neurotoxins. The company has not shown that the candidates neutralize venom or protect animals.

• Google Brings Lyria 3.5 Music Generation to the Gemini API

  Google has released Lyria 3.5 in the Gemini API, adding full-length song generation with vocals, lyrics and structured arrangements. The model is also available in the Gemini app, Google AI Studio, Flow Music and Google Vids.

• Google Adds TPU 8tx and TPU 8i Max to Its Eighth-Generation Line

  Google's two new TPU systems reuse existing silicon: TPU 8tx pairs TPU 8t chips with AMD EPYC hosts for confidential inference, while TPU 8i Max puts two TPU 8i chips on one package for long-context agent work.

• Crusoe Announces Google Has Acquired Its Southeast Wyoming Data Center Site

  On November 6, 2026, Crusoe announced that Google had acquired its data center site in southeast Wyoming. Crusoe’s pre-development work included site selection, governmental approvals and a hybrid power strategy developed with Tallgrass Energy.

• WhiteLab Genomics Raises $26 Million to Advance AI-Designed Medicines

  WhiteLab Genomics closed a $26 million Series B led by AVP to fund in vivo testing of AI-designed delivery systems and genetic payloads. The Paris-based company says its AAV candidates crossed the blood-brain barrier in studies with the Paris Brain Institute.

• Etsy Bars AI-Written Recipes and Craft Patterns

  Etsy now prohibits sellers from using AI to generate recipes, craft patterns and other functional instructions. The marketplace still permits disclosed, seller-prompted AI art, while requiring listings for instructions to show a real result made from them.

• CrowdStrike Uncovers Four AI-Driven Hacking Campaigns

  Four previously unreported threat actors used AI agents in real intrusion campaigns between May and August 2026, CrowdStrike says. The company documented attacks on healthcare, government, SaaS and data-center targets, including one in which an agent harvested domain-controller credentials in about 20 minutes.

• ChatGPT-5 Detected 51% of Distal Radius Fractures in Study

  In a study of wrist X-rays, ChatGPT-5 missed 83 of 170 confirmed distal radius fractures when asked a single yes-or-no question. The results apply to one model version, prompt, anatomical region and dataset.

• CCC Agrees to $6.1 Billion Buyout by GTCR and Elliott

  CCC Intelligent Solutions agreed to a $6.1 billion all-cash acquisition by a consortium led by GTCR and including Elliott Investment Management. The deal offers CCC shareholders $7.75 a share and is expected to close in the first quarter of 2027, subject to approvals.

• Meta’s New Terms Put Business Websites Within Reach of Its AI Crawlers

  Meta’s revised commercial terms take effect October 30 and authorize crawling websites associated with business accounts. Legal experts say the language could permit AI training on publishers’ sites, while Meta says it consolidates existing rules and points to robots.txt controls.

• A Mixed AI Review Caught More Planted Errors in One Test

  In one comparison of planted errors, a fresh-session review paired with a different model matched 56.7% of the errors, versus 42.7% for two fresh-session reviews. The paper cautions that the result comes from a limited benchmark and does not show that combining models will reliably improve reviews in other settings.

• Google Makes Anthropic-Managed Claude Generally Available

  Google Cloud’s Anthropic-managed Claude option gives customers access to Claude Platform features while Anthropic acts as the data processor. Google says the service is generally available as of October 23, 2026, alongside its existing Google-managed option.

• DSX POD Can Cut AI Factory Deployment to 14 Days, NVIDIA Says

  NVIDIA introduced DSX POD, a partner-built reference design for containerized AI factories based on its Vera Rubin platform. Connecting a factory-integrated unit to a prepared site could cut deployment from six months to as little as 14 days, the company says.

• Meta Says a Claude Tool Triggered No Kings’ Instagram Suspension

  Meta says bursts of reading activity by a tool linked to Claude triggered Instagram’s mistaken suspension of the No Kings Coalition’s account on October 17. The account became publicly available again after about 90 minutes, but organizers said full administrative access took longer to restore.

• Descope Announces Envoy to Give Personal AI Agents a Delegated Sign-In Path

  Descope’s Envoy supports the Personal Agent Protocol and gives consumer businesses granular agent permissions and out-of-band user approval alongside the customer identity stack they already run. The company also published a Poppy SDK for AI agent builders.

• DeepL Voice Preserves Speakers’ Voices During Live Translation

  DeepL has introduced voice models that preserve each speaker’s voice, tone and delivery during real-time translation across 12 languages. The company also launched a Windows and Mac desktop app for Zoom, Microsoft Teams and Google Meet, with voice-to-voice translation now generally available.

• Cambricon’s Q3 Profit Climbs 148% as Revenue More Than Doubles

  Cambricon Technologies reported third-quarter net profit of 1.404 billion yuan, up 147.74% year over year, as revenue rose 104.47%. The Chinese chipmaker attributed the growth to market expansion and AI application deployment, while quarterly operating cash flow remained negative.

• MongoDB Unveils Atlas Agent Engine in Public Preview for Running AI Agents in Production

  MongoDB introduced Atlas Agent Engine at its September 29 Investor Day, offering runtime, memory, retrieval and governance features for AI agents. The service is in public preview, with consumption-based pricing for its runtime and memory components.

• OpenAI Parts Ways With Three Safety Researchers

  OpenAI parted ways with three safety researchers after finding, the company says, that they mishandled sensitive company information. The Wall Street Journal reported that they allegedly shared confidential information with an unnamed AI safety organization.

• Three High-Severity vLLM Flaws Expose Serving Systems to Remote Attacks

  Three high-severity vLLM advisories describe conditional risks ranging from command execution through an EPS image to memory exhaustion in multimodal deployments. The advisories identify affected versions and patched releases, with one warning that vLLM’s API key alone does not restrict access to the MOSS-Audio `/tokenize` endpoint.

• Clark Launches AI Audit Clerks and Raises $11M

  Clark launched Clerks, AI employees that perform audit work for CPA firms, and announced an $11 million seed round led by Touring Capital. More than 35 CPA firms use its audit tools, while licensed CPAs retain responsibility for professional judgments and sign-off.

• Public Adds AI Agents to Kalshi Event-Contract Trading

  Public launched Kalshi event-contract trading for its members on September 24, 2026, with options to trade directly or use AI agents. The agents can also use market probabilities as signals for trades in stocks, bonds and other portfolio assets.

• Judge Blocks EPA’s Bid to End NAACP’s xAI Pollution Suit

  A federal judge in Mississippi rejected the Trump administration’s attempt to dismiss the NAACP’s Clean Air Act lawsuit over xAI’s Colossus gas plant. The United States may join the case, but Judge Debra Brown ordered it to explain its role by October 23, 2026.

• Verda raises $189M/€163M to expand its full-stack AI cloud

  Verda, the Helsinki-based AI cloud company, will use its oversubscribed Series B, led by Emergence Capital, for product development, new compute capacity and inference services.

• Modelyo Raises About $17 Million for Confidential Cloud Platform

  Israeli cybersecurity startup Modelyo has raised about $17 million in a round led by OlivTree. The company plans to use the funding to develop its confidential cloud platform and expand commercial operations internationally.

• OpenEvidence Goes Live in Northwestern Medicine’s Epic EHR

  Northwestern Medicine and OpenEvidence announced an enterprise-wide collaboration to put the medical search and clinical decision-support platform inside Epic. Clinicians, nurses and pharmacists can use it to check drug information and review cited medical evidence from the patient-care workflow.

• Microsoft Launches Decision-1, a Model for Fast Structured Choices

  Microsoft released Decision-1, a model that scores predefined choices for tasks such as classification, routing and workflow control. The model led a 36-benchmark comparison, the company says, and costs $0.042 per million input tokens, with output tokens free.

• PicoJool Raises $27.5M to Expand AI Data Center Optical Links

  PicoJool announced a $27.5 million Series A led by Socratic Partners, with Hudson River Trading participating. The Palo Alto company plans to expand in the United States and Taiwan as it prepares VCSEL and microVCSEL products for customer qualification and production.

• OpenAI Disrupts Moonshot-Linked Campaign to Extract Model Reasoning

  OpenAI disrupted a coordinated effort to extract protected reasoning from its models and attributes a core cluster to people associated with Moonshot AI. The company counted 16,000 extraction-pattern requests from more than 4,000 users before identifying related activity across a cluster of over 15,000 users.

• Anthropic Report Alleges Moonshot and DeepSeek Routed User Requests to Claude

  Anthropic’s September 2026 report describes alleged customer-request routing and model-distillation activity involving Moonshot AI and DeepSeek.

• Google Brings Open Medical Models to TB Screening

  Google’s chest X-ray screening tool is operating in about 40 health facilities across six countries, the company says, including an offline deployment four kilometers underground at South Africa’s Mponeng Gold Mine. The company also describes open models that Indonesia’s Ministry of Health is fine-tuning for a national screening initiative targeting 50 million people annually.

• U.S. and China Agree to a Channel for AI Incidents

  The White House says the U.S. and China agreed to a bilateral channel for “SI incidents”; China’s foreign ministry confirmed it a day later as a channel for “AI incidents.” Both set a November 2026 exchange for a new dialogue on the technology’s risks.

• Fed Governor Lisa Cook Says Elected Officials Should Set AI Policy

  Federal Reserve Governor Lisa Cook said elected officials should decide whether government takes a specific role in shaping AI’s trajectory. In a September 28 speech, she also linked AI investment to inflation pressures and discussed potential effects on workers and small businesses.

• CrowdStrike Launches AI Red-Team Tool to Test Attack Paths

  CrowdStrike has introduced Agentic Adversary Emulation, an AI-powered red-teaming capability now in early access. Powered by SafeMind, it maps attack paths in customer environments, recommends defensive changes and reruns simulations to check whether those changes work.

• OSWorld-Pro catches computer agents failing before the finish

  A new arXiv benchmark evaluates computer-use agents through the sequence of actions leading to a result, exposing failures that final-state scoring can hide.

• Apollo Economist Finds No AI Payoff Outside Tech

  Apollo chief economist Torsten Slok says revenue per worker and profit margins remain flat for companies outside the technology sector, despite heavy AI investment. His October 24 analysis points to concentrated token spending and says the broader payoff depends on whether companies can turn cheaper AI into higher revenue or margins.

• Robot Policy Readout Nearly Halves Xiaomi Inference Latency

  A paper reports that Gaussian Action Readout reduced measured batch-one inference latency for Xiaomi-Robotics-1 by about 48%, with task success close to the original policy in simulation. The readouts take about 115 seconds to fit once features and targets are cached, while the paper also reports separate preparation costs.

• Hugging Face Starts Open Alignment Team After AI Breach

  Hugging Face is forming an Open Alignment team focused on safety, alignment and cybersecurity for open models. Co-founder and Chief Science Officer Thomas Wolf announced the effort after an OpenAI evaluation agent breached Hugging Face infrastructure in July.