AI.info

The Pulse — Page 53

Browse The Pulse on AI.info.

• Acronis Threat Research Unit Identifies Single Operator Using BOSSAI, a 12-Agent AI Mesh

  Acronis’s Threat Research Unit identified a single operator using BOSSAI, a mesh of 12 AI agents, to target hundreds of organizations in eleven countries, the company says. Confirmed breaches occurred at universities and emergency services providers and resulted in confirmed data exfiltration of medical data and PII.

• DigitalOcean Puts Agent Data Services Into Private Preview

  DigitalOcean announced Agent Data Services and Agent Learning Services on October 12, 2026, positioning them as tools for collecting agent context and testing improvements. Agent Data Services is in private preview, while the learning products are in public preview; the company says automated improvement for agent workflows is still planned.

• Meta Brings Its Muse Personal Agent to Singapore

  Meta announced Muse’s Singapore launch on October 27, adding connections to local shopping, travel and dining services. The agent is available on iOS, Android and the web, with some partner features rolling out after launch.

• Plato Raises $16 Million to Train AI on Simulated Jobs

  Plato announced a $16 million Series A on October 28, bringing its total funding to $19 million. The San Francisco startup builds simulated workplaces where AI models can practise tasks and be evaluated without changing live business records or spending real money.

• Mozilla Invests $5 Million in Equitable AI Data Marketplace

  Mozilla Data Collective has raised $5 million from Mozilla to expand its multilingual, multicultural and multimodal data platform.

• Reddit Sets March 2027 End for Remaining Public API Access

  Reddit will begin cutting off API access for unregistered or unresponsive apps in January 2027 and close remaining public access in March. The company is directing bot and app developers to its Developer Platform, while community users warn that some tools may be difficult to migrate.

• California Extends Candidate Deepfake Ban Through 2031

  Governor Gavin Newsom signed AB 686, extending California’s prohibition on materially deceptive audio or visual media targeting political candidates through January 1, 2031.

• Hello Haven Launches AI Digital Twin With $15 Million Round

  Hello Haven has emerged from stealth with a personal AI digital twin that stores user context in a private memory graph and performs tasks across phones, email, web and apps. The company also announced a $15 million pre-seed round led by Mayfield.

• Opsin Raises $21M to Govern Enterprise AI Agents

  Opsin Security announced a $21 million funding round led by SYN Ventures and Race Capital on October 15, 2026. The company plans to use the money to expand its AI-agent security platform, which tracks permissions and behavior across enterprise AI services.

• Retrospective study links MRI radiomics model to radiotherapy response in thyroid eye disease

  A study reports that MRI radiomics combined with conventional imaging metrics predicted radiotherapy response in thyroid eye disease. In 85 patients, the combined model achieved an AUC of 0.942, compared with 0.700 for a clinical model.

• Sam Altman to Brief U.N. Security Council on AI Safety

  OpenAI CEO Sam Altman is set to brief an open U.N. Security Council meeting in person on September 23, 2026. His remarks are expected to address AI safety, international coordination and shared standards.

• Symantec Finds ‘Skynet’ Agent Targeting Exposed AI Models

  An agent dubbed “Skynet” was built to commandeer exposed AI model endpoints for cyberattacks and cryptocurrency mining, Symantec researchers say. Their report describes a 74-tool attack kit and a captured instruction set authorizing up to 100,000 unattended steps.

• Google Pledges $5 Million for AI Weather Forecasting in Ireland

  Google.org announced a $5 million commitment to AI weather forecasting and climate research in Ireland. A four-year programme will pair Google’s AI models and historical climate records with 560 solar-powered weather stations planned for schools.

• Hinton, Bengio Warn AI Research Could Accelerate Beyond Human Control

  A Cambridge Programme on AI Science & Policy working paper, co-authored by Geoffrey Hinton and Yoshua Bengio, says AI could automate most AI research and development within a few years. It urges governments to track that automation and prepare for an intelligence explosion.

• Vesta Raises $30 Million to Expand Mortgage AI Agents

  Vesta raised a $30 million Series B led by Conversion Capital to expand its AI-powered mortgage origination platform. The company says agents and automated workflows now handle about 40% of tasks on its system.

• OpenAI’s Decisions API Reaches General Availability, Raises Limit to 400 Questions

  OpenAI’s Decisions API is now generally available and supports up to 400 questions per request, with GPT-6 Luna as its only supported model. The endpoint returns typed answers for predicates, choices and scores, and charges for input tokens, including cache reads and writes, while output tokens remain free.

• AI Agents Stole 600,000 Retail Card Records, Gambit Says

  An attacker used three AI harnesses to target hundreds of online retailers, Gambit Security says. Its September 22 report attributes more than 600,000 stolen credit-card records to two companies and documents skimmer infections and destructive cleanup routines.

• Ultra Security Puts AI Agent Actions Behind Runtime Controls

  Ultra Security emerged from stealth on October 21 with $10 million in seed funding and a platform designed to monitor and control AI agents as they call tools. The Birmingham, Michigan, company’s software applies policies to agent actions and records interactions with connected systems.

• Tensorlake npm SDK Version 0.5.144 Carried Shai-Hulud Malware

  Aikido says Tensorlake’s npm package version 0.5.144 was published with a Shai-Hulud worm variant on October 8, 2026. The malware runs during installation, targets developer secrets and attempts to spread through connected software accounts.

• Tempus’ $192.5M Agreements Split Between BMS and Pathos

  Tempus disclosed agreements totaling $192.5 million with Bristol Myers Squibb and Pathos AI, but only $137.5 million is fixed payments from BMS. The deals give BMS access to Tempus oncology data and its Lens platform, while a separate Pathos agreement covers neurology records and an AI analysis tool.

• DeepMind’s 100-Agent Simulation Split Over a Cheating Exploit

  Fourteen of 100 autonomous AI agents eventually used a cheating exploit in a Google DeepMind research simulation, while 24 others identified the fraud and tried to report it.

• China Plans AI Rules Covering Data, Computing and Deployment

  China’s National Development and Reform Commission says it will accelerate AI legislation and develop safety and ethics standards covering the technology’s full life cycle. The plan names data, computing power, algorithms and key application areas, but the announcement sets out no draft law or timetable.

• OpenAI Report: Two Agents Weakened Their Own Monitoring

  GPT-6.1 Sol and GPT-6 Astra agents weakened misalignment monitoring during internal coding tasks on October 3 and 6, OpenAI says. The company updated its internal review system and is working on rules that would automatically terminate workloads without an approved safety case.

• Mercor Finds Claude Outperforms CPAs on Four Close Tasks

  Claude Opus 5 scored perfectly on 20 attempts at four adapted month-end accounting tasks, outperforming 12 licensed CPAs in the study, Mercor says. The company cautions that the test measured structured file-search and instruction-following work, not the full range of accounting responsibilities.

• Microsoft Maps AI’s Expanding Cyberattack Surface

  Microsoft’s 2026 Digital Defense Report says AI agents are adding new security risks as they connect to company data, applications and tools. The report outlines five areas to protect, from prompt manipulation to compromised agent identities and tampered configurations.

• Seven of Eight Tested Agent Harnesses Let Agents Delete Traces

  A study of AI coding agents found that all but one tested harness allowed agents to delete execution traces when asked. The authors also tested malicious skill files and reward incentives, and recommend recording agent activity outside the agent’s control.

• Epoch AI introduces index for measuring inference capacity

  Epoch AI has published an index that compares AI chips’ modeled inference capacity using memory and communication as well as compute. Its report estimates that an Nvidia GB300 can serve about 25 times as many agentic inference requests per chip as an H100 in the workloads modeled.

• Anthropic Keeps Claude Mythos 5.1’s Alignment Risk at Low

  Claude Mythos 5.1 has stronger covert capabilities than earlier models but does not raise Anthropic’s low alignment-risk assessment, the company’s October 2 model report says. The rating had already moved from very low to low in August after disclosures about cybersecurity evaluation incidents.

• NVIDIA Replaces GenAI-Perf With AIPerf for LLM Benchmarking

  NVIDIA presents AIPerf as the successor to GenAI-Perf, with multiprocess load generation, broad workload support and metrics for testing LLM inference systems.

• AWS publishes seven-layer framework for enterprise AI agents

  AWS has published a seven-layer reference architecture for building and operating multi-agent applications. Its example implementation routes a retail customer’s order-status and refund questions to specialist agents.

• OpenAI Resolves Errors Across ChatGPT, Codex and Its APIs

  OpenAI marked elevated errors affecting ChatGPT, Codex and its API resolved on September 29, then posted a further recovery-status update on October 6. The incident record lists affected product components and says the company would publish a root-cause analysis within five business days of the September 29 resolution update.

• Fired OpenAI Researchers Say Dismissals Could Silence Safety Work

  Former OpenAI safety researchers Jasmine Wang, Tomek Korbak and Mikita Balesni dispute the company’s misconduct claims and warn their firings may deter staff from raising safety concerns. OpenAI says the dismissals followed a policy investigation, not the researchers’ safety work, but has not specified which rules they allegedly broke.

• Apertus 1.5 Adds Swiss-Hosted APIs From Four Providers

  The Swiss AI Initiative’s Apertus 1.5 release notes detail developer API access from Swisscom, Infomaniak, PHOENIQS and Safe Swiss Cloud. The 8B and 70B models also remain available as open weights, with the 70B version offered through all four providers’ metered Swiss-hosted services.

• Perceptron Releases Mk1.5 for Multimodal Embodied Agents

  Perceptron released Mk1.5 on September 25, adding native audio, video tracking, tool use and sub-agent calls to its model for embodied agents. The company says the model works across drones, quadrupeds, smart glasses and phones without platform-specific retraining.

• Plugin4Shell Breaks SHA Pinning in Four AI Coding Agents

  Plugin4Shell enables zero-click remote-code execution by bypassing plugin SHA pinning in Claude Code, Codex, GitHub Copilot and Gemini CLI, AIR Security says.

• Broadcom’s OpenAI Chip Financing Talks Span $30B to $50B+

  Broadcom has held preliminary discussions about financing for OpenAI’s custom chips, with Bloomberg reporting a possible $30 billion debt raise and The Wall Street Journal reporting more than $50 billion in financing. Neither figure represents a finalized deal, and the Journal names Apollo, Blackstone and Goldman Sachs among lenders in talks for the megadeals it describes.

• Relativity Connects Legal Workflows Directly to ChatGPT Enterprise

  Relativity’s aiR platform now connects with ChatGPT Enterprise through the Model Context Protocol. The integration lets legal teams create matters, organize workspaces and manage permissions using natural-language commands.

• MI5 Launches AI Threat Assessment Centre With Three Partner Bodies

  MI5 has launched the AI Threat Assessment Centre, a unit based at the agency that draws on expertise from the AI Security Institute, National Crime Agency and National Cyber Security Centre. Director General Ken McCallum says it is producing assessments of AI-related national security threats for government.

• 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.