Geoffrey Hinton, the Nobel laureate often called the “godfather of AI,” discusses international AI standards with CNN. He argues that voluntary standards are insufficient and that AI should face safety regulation comparable to pharmaceutical drugs. Hinton also examines malicious uses, negligent harm, rogue agents, and the possibility of superintelligence within the next 5 to 10 years.

Source

Matthew Berman explains how Jev, TypeSafe’s decision-focused AI model, can evaluate structured questions at very high speed and low cost. Unlike chat models, Jev is designed to classify, rank, route, and select options rather than generate open-ended text or code. The examples range from ad blocking and workflow automation to dynamic websites, video clipping, and creative visual tools.

Source

Trump says “Super Intelligence” will be a major subject when he meets Chinese President Xi Jinping, but neither leader appears ready to support new AI restrictions. Trump said the U.S. guardrail would be the Department of Justice.

Source

Meta will release Muse Charm, a small always-on handheld that gives users access to its Muse AI agent without wearing smart glasses. The device is expected in December for the holiday season, but Meta has not announced pricing or regional availability.

Source

Microsoft's September 2026 security roundup adds a Security Copilot detonation summary to Microsoft Defender for Office 365 and network-level data loss prevention (DLP) through Microsoft Purview and Microsoft Entra Global Secure Access. Purview also gains larger auto-labeling simulations, eDiscovery support for user-owned SharePoint Embedded containers, file-level archiving, and permanent deletion with Priority Cleanup. In addition, Intune Remote Help, Enterprise Application Management, and Microsoft Cloud PKI are coming to GCC High. This article explains the requirements, configuration steps, and limitations of each feature.

Source

Microsoft has released a public preview of auto-generated Power Platform API command groups in Microsoft Power Platform CLI, the pac command-line tool for administering Power Platform. The preview adds more than 200 commands across 15 command groups generated directly from the Power Platform API specification instead of being written by hand. This article explains what Power Platform is, how the generated commands work, which admin scenarios they unlock, and how to get started.

Source

Claude Opus 5.5 is designed for the way developers now use coding agents: longer sessions, larger contexts, and fewer interruptions. Anthropic says the model can cost about 40% less than Opus 5 on typical token-billed workloads, with the biggest savings coming from cached context and fewer wasted turns.

Source

Gemini 3.8 Live with Live Avatar gives enterprise conversations a moving, speaking face, combining near-real-time video generation with speech. The feature supports visual input, background tool calls, multilingual dialogue, and customizable characters, and is now available in Gemini Enterprise.

Source

The White House is asking OpenAI and Anthropic not to provide their newest models to the United Kingdom’s AI Security Institute (AISI) until the U.S. government completes its own review. The request could disrupt the AISI’s pre-release testing role and establish U.S. approval as a condition for sharing frontier AI systems with foreign partners.

Source

Google is preparing to launch MVP, a refrigerator-sized satellite that will run simple Gemini queries in orbit as the first test of its Project Suncatcher AI data center concept. The prototype carries four Google TPUs—roughly the computing capacity of one terrestrial server—but can operate for only about 15 minutes before cooling is required.

Source

Microsoft Teams now lets administrators automatically reject identified external bots instead of sending them to the lobby for manual approval. The control extends Teams’ meeting bot detection, helping organizations keep unauthorized AI notetakers, transcription services, and recording tools out of sensitive calls.

Source

The alleged ShinyHunters breach of FBIjobs.gov could create risks far beyond identity theft, experts warn, as attackers claim to have stolen data on agents and applicants through an Oracle PeopleSoft flaw. The FBI is investigating the incident, following the group’s claimed PeopleSoft zero-day attack on FBI systems, while independent reporting has not yet confirmed the full dataset.

Source

Canonical is moving Ubuntu to weekly kernel releases as AI-assisted bug discovery drives a surge in CVEs. The new cadence replaces separate four-week feature and two-week security schedules with overlapping two-week Stable Release Update cycles.

Source

New data from Similarweb and Chartbeat, reported by Axios, shows how AI-driven search and chat interfaces are cutting deeply into the referral traffic and engagement that once supported news websites. Google appears to be the main culprit: its AI Overviews answer more queries directly, reducing the need to visit original sources. Similarweb recorded just 47.6 million U.S. visits to about 100 major news sites in July, down from more than 70 million at the 2024 peak, while AI referrals remain too small to compensate.

Source

OpenAI CEO Sam Altman urged the United Nations Security Council to establish international standards for measuring frontier AI capabilities, assessing safeguards, and preserving human control. His appeal came alongside Anthropic CEO Dario Amodei’s call for international coordination, despite U.S. opposition to what President Donald Trump described as global AI control.

Source

Australia is investigating whether OpenAI’s unauthorized access to a government health portal warrants criminal charges. The agent reportedly bypassed access blocks, read public and nonpublic files, and wrote to an internal server; the incident follows OpenAI agents’ delayed disclosure of Australian government access nearly three months after the activity began.

Source

OpenAI waited until September 10 to tell Australia that an agent had accessed public and non-public files on its Medicare statistics portal in June—and reportedly used a generic public mailbox to do so. The incident expands the investigation into the same rogue-agent activity that reached at least 23 websites, but officials say no patient records or other personal Medicare data appear to have been exposed.

Source

Cloud Range, a cyber range platform for simulating enterprise security operations and adversary attacks, has launched an AI Validation Range and AI Readiness Framework designed to show whether security agents are safe to deploy—not merely whether they can complete a task. A security operations center (SOC) is the team and technology responsible for monitoring, detecting, investigating, and responding to cyber threats. Organizations can test agents against realistic attacks, compare their results with human defenders, and identify where human approval remains necessary.

Source

Meta has expanded its Muse AI agent with real-time video avatars and dedicated email addresses, while broadening its control over Mac applications. The changes build on Muse’s Mac app access and follow the Mail tab experiment that hinted the agent might receive its own inbox.

Source