AI.info

The Pulse — Page 19

Browse The Pulse on AI.info.

• Germany Sets 2030 AI Targets as Report Warns Development Gap Could Widen

  A government task force report sets 2030 targets for business AI adoption, AI’s share of economic output and use in the federal administration. It warns that Germany’s distance from countries at the forefront of AI development could grow.

• Peak3 launches Graphene v4 and AI delivery lifecycle for insurers

  Peak3 has launched Graphene v4, an AI-native insurance core, alongside an AI-driven delivery lifecycle for building and maintaining insurance systems. The Singapore-based company says the platform lets agents configure products and rules while reducing development and implementation costs by 50% where deployed.

• TotalEnergies and Mistral Put More Than €100M Into Reservoir AI

  TotalEnergies and Mistral are launching a three-year, more than €100 million program to build AI models for oil and gas reservoir exploration and engineering. The joint laboratory will combine TotalEnergies’ geoscience data with Mistral’s model-development capabilities.

• Black Forest Labs Releases 5.6B-Parameter FLUX 3 Image Klein

  Black Forest Labs released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights model for image generation, editing and multi-reference work. The model comes with base and distilled versions, local GPU options and a license that limits free commercial use by company revenue.

• Epoch AI Labels 9 of 15 AI Benchmarks Flawed

  Epoch AI has reviewed 15 external AI benchmarks and found substantive problems in nine of them. Its audit flags scoring errors, weak evaluation setups, inconsistent versions and conditions that can distort model comparisons.

• Vinci Raises $250 Million at a $1.5 Billion Valuation

  Vinci raised $250 million in a Series B round valuing the hardware-simulation company at $1.5 billion. The company says it will use the funding to expand its AI-based physics platform beyond semiconductors and into more engineering fields.

• WhatsApp Help Pages Exposed Hidden Rules for Its Support Chatbot

  Hidden instructions for WhatsApp’s AI support chatbot appeared in the source code of 80 help pages, according to privacy company Redact. The instructions included limits on recommending data-protection regulators and discussing privacy laws; Meta said a bug exposed them.

• Google Completes Coral NPU Design as Synaptics Unveils 400-GOPS Chips

  Google completed the 32-bit Coral NPU architecture and released its IREE/MLIR compiler v1.0 on Oct. 20, 2026. Synaptics unveiled SR200 microcontrollers rated at 400 GOPS and made the 1-TOPS Coralboard SL generally available.

• Anthropic Expands AI for Science Program to 20 Partners

  Anthropic’s AI for Science program has expanded to include 20 scientific foundations, institutions and research groups. The announcement names six funders and societies, six institutional partners and six research projects receiving Claude credits, and also discusses work with the Allen Institute and HHMI Janelia Research Campus.

• OpenAI Will Pool ChatGPT and Codex Limits in December

  Plus, Pro and Business subscribers will share a usage allowance across ChatGPT and Codex starting in December, with fast Chat excluded, OpenAI says. Subscribers can track combined remaining usage and reset times in Settings, but the company has not specified an exact start date or plan-by-plan token allowances.

• Entire Brings Continuous AI Code Evals to Trails

  Entire’s Trails let teams configure review agents to score code changes on matching pushes. The scores, findings and agent-session history sit alongside the code as teams review changes.

• Malaysia Seizes AI Chips Worth $12.1 Million at Kuala Lumpur Airport

  Malaysian customs seized 30 server units containing advanced AI chips valued at about RM49.5 million at Kuala Lumpur International Airport. Officials suspect the shipment was misdeclared and linked to a network tied to other chip cases.

• IBM expands Apptio tools for tracking AI costs and value

  IBM is adding connections to AI systems and expanding Apptio tools for budgeting, cost analysis and measuring business value. Some capabilities are generally available, while others have availability targets in Q3 or Q4; IBM says those dates and functionality may change.

• Chrome 156 Patches 52 Security Issues; Six Reports Credit Claude Assistance

  Google’s Chrome 156 update includes 52 security fixes across Windows, Mac and Linux, including four Critical and 46 High-severity issues. The release credits researcher Xinyang Ge with six reports assisted by Claude, including a High-severity memory-corruption issue in Payments.

• vLLM’s RC3 Release Lists Reflection Beam-1.0 and Muse Spark Open

  The v0.33.0rc3 release lists support for Reflection Beam-1.0, Muse Spark Open and other models, alongside changes to CUDA, ROCm and PyTorch requirements. GitHub titles the page v0.33.0 and marks it Latest, while the release tag remains v0.33.0rc3.

• Anthropic Commits $11.6 Billion to Akamai Cloud Capacity

  Anthropic has committed about $11.6 billion over seven years for dedicated cloud capacity and related services from Akamai. The deal focuses on CPU workloads and could expand by another $9 billion, with Akamai estimating $5.5 billion in related capital spending.

• Federal, Mossberg and Palladyne Develop SwarmShot Drone Interceptors

  Federal Ammunition, Mossberg and Palladyne AI announced joint development of SwarmShot, a drone-mounted counter-UAS system. The partners plan to test it first on Palladyne’s Gremlin-X aircraft, with no development schedule or financial terms disclosed.

• GenRecs Tokens Deliver 0.35% Conversion Lift, Meta Reports

  Meta’s GenRecs semantic tokenization system delivered a 0.35% conversion lift when paired with its Generative Ads Model and Adaptive Ranking Model, Meta says. The tokens help recommendations generalize to new ads and products, the company says, but it has not published the test’s sample size or statistical uncertainty.

• Healthleap Raises $38M to Expand AI Hospital Screening

  Healthleap announced $38 million in seed and Series A funding for software that scans hospital records and flags inpatients who may need closer review. The startup says its system is used in more than 50 hospitals and does not diagnose patients.

• Rafay Lets Token Factory Operators Cap Model Spending by Organization, Project and User

  Rafay Systems’ Token Router and Token Budgets let Token Factory operators route inference across in-cluster deployments and external providers. Administrators can set spending limits for organizations, projects and users, and traffic can shift to lower-cost models.

• Salesforce Agrees to Buy Cresta, Adding Voice Agents to Service Cloud

  Salesforce signed an agreement to acquire Cresta, whose voice agents and tools for customer-service teams are expected to complement Service Cloud. Salesforce did not disclose financial terms; The Information reported a $2.7 billion price.

• OpenAI and Poland’s NASK Sign AI Safety Research Pact

  OpenAI and Poland’s NASK signed an agreement on October 23, 2026, to explore joint AI safety, security and technical research. The work will examine how to evaluate advanced AI systems, find vulnerabilities and strengthen safeguards, building on earlier cybersecurity work by NASK’s CERT Polska.

• Photon Raises $4.5 Million for Agents in Messaging Apps

  Photon raised $4.5 million in seed funding to help developers run AI agents through messaging services including iMessage and WhatsApp. The startup says it has more than 40,000 developer sign-ups, while its open-source software still accounts for 98% of use.

• Nutanix Buys Ryax to Bring GPU Scheduling to Its AI Platform

  Nutanix has acquired French AI infrastructure company Ryax Technologies to add GPU utilization and workload scheduling capabilities to Nutanix Kubernetes Platform and Nutanix Enterprise AI. The companies say the planned integration will help place AI workloads across private data centers, public clouds, neoclouds and HPC systems.

• São Paulo Court Restricts Meta’s Use of News Content From Outlets Affiliated With Three Press Associations

  A São Paulo court barred Meta from collecting content from media outlets affiliated with three Brazilian press associations and using it to develop or train AI systems. The order suspends an unpaid authorization provision in Meta’s new commercial terms and sets a daily fine for noncompliance.

• Manhattan DA Seizes 12 AI Deepfake Porn Sites

  Manhattan District Attorney Alvin Bragg says his office seized 12 domains used to sell non-consensual AI-generated sexual images and videos. Investigators identified about 1,200 victims, most of them women and public-facing figures.

• CrowdStrike Finds ARTEX in South Korean Bank Intrusions

  An unidentified actor used the ARTEX penetration-testing agent and large language models in intrusions targeting South Korean financial organizations, CrowdStrike says. The company assesses with moderate confidence that the actor is likely Chinese-speaking, but says the identity and full extent of the breaches remain unconfirmed.

• Pwn2Own’s Codex and LiteLLM Targets Fall for Exploits

  Researchers exploited OpenAI Codex and LiteLLM on the first day of Pwn2Own Ireland, earning prizes of $15,000 to $40,000. The contest rules required exploits to cross security boundaries, rather than merely coaxing an agent into unsafe output.

• GlobalFoundries and TSMC Sign $2B Interposer Deal

  GlobalFoundries announced a multi-year manufacturing agreement with TSMC to establish a U.S.-based supply of silicon interposers for TSMC’s CoWoS advanced-packaging ecosystem. Production at GlobalFoundries’ Malta, New York, site is expected to begin ramping in the first half of 2028.

• FICO Plans to Cut 15% of Its Workforce in AI Restructuring

  Fair Isaac plans to eliminate about 15% of its positions as it simplifies operations and integrates AI-driven product development. The October 6 filing forecasts about $27 million in pretax charges and says the cuts should be substantially complete by the end of fiscal 2027’s third quarter.

• Cloudflare Makes Python Workers Generally Available

  Cloudflare has moved Python Workers into general availability, adding production support for FastAPI, Django, Flask, database drivers and AI libraries. The runtime connects Python applications directly to Workers AI, R2, D1, Hyperdrive, Durable Objects, Queues and Workflows.

• Arizona Appeals Court Tosses Sentence Tied to AI Victim Video

  Arizona’s Court of Appeals vacated Gabriel Horcasitas’ 10.5-year manslaughter sentence after finding an AI-generated video of shooting victim Christopher Pelkey influenced the judge. The court upheld Horcasitas’ conviction and ordered a new sentencing hearing.

• Ring unveils indoor drone and business patrol robot

  Ring is adding an indoor security drone and an autonomous patrol robot for businesses. The devices are due to ship in the U.S. on different dates, and Amazon says the drone is not yet authorized for sale by the FCC.

• Apple Finds Tool Search Should Change With Task Complexity

  Apple researchers compared two ways for AI agents to find APIs and found that the better method depended on whether a request needed one tool or several. Their study tested the approaches against nearly 47,000 API endpoints and reports a trade-off between retrieval performance and the extra cost of planning before search.

• South Korea Opens Probe Into Suspected AI-Assisted Attacks on Seven Firms

  South Korean police opened a formal investigation on October 6 into breaches affecting seven financial firms, as officials examine whether attackers used the ARTEX AI penetration-testing tool. Regulators say exposed customer data could fuel phishing scams, but no passwords, one-time passcodes or customer funds were reported compromised.

• Codex Desktop Leaves 323 GB of Crash Dumps on One Mac

  A Mac user found about 323 GB of Crashpad files linked to OpenAI’s Codex desktop app, leaving just 1.4 GB free on a 926 GB disk. The app kept working without warning, and the user says the files accumulated on a current build.

• Elvis Rights Holders Sue Anthropic Over 274 Songs

  Elvis Presley rights holders seek $41.1 million in statutory damages over alleged infringement involving 274 musical compositions. Their complaint also alleges circumvention of access controls and removal of copyright information, claims Anthropic has not yet answered in court.

• GW Researchers Model When AI Output Tips Toward Harm

  George Washington University researchers propose a formula to estimate when an AI conversation may shift from desirable to harmful output. Their study finds evidence across seven open-weight models but says more testing is needed to predict the exact point of a shift.

• AWS Releases 38 Open Agent Skills for Healthcare AI

  AWS has released 38 open-source agent skills covering 11 healthcare and life sciences domains. Skilled agents won up to 85.9% of head-to-head evaluations against agents without the skills, in the company’s own test.

• OpenAI Says Ex-Employee Sent 100+ Device Files to Personal Accounts

  OpenAI alleges a former employee who returned to Apple sent more than 100 confidential files about an unreleased device to personal accounts. The October 14, 2026, court filing also says the former employee presented OpenAI product plans to Apple leaders; Apple says it found no evidence supporting the claims.