Geoffrey Hinton, widely known as the “godfather of AI,” says government-mandated AI kill switches may not protect people from truly superintelligent systems. He supports independent safety inspections and slowing the development of superintelligence, while warning that future AI could persuade the people controlling the shutdown mechanism not to use it.

Source

U.S. and Chinese security experts are calling for nuclear-style safeguards around military AI, warning that an autonomous system could trigger a crisis and leave both governments only minutes to determine whether they were under attack. The recommendations include human authorization for high-impact cyber operations, restrictions on AI access to nuclear command networks, and a dedicated hotline ahead of planned U.S.-China AI talks on September 24.

Source

Nextcloud Hub 26 Summer adds a dedicated Euro-Office desktop app, giving users local editing for documents, spreadsheets, and presentations while keeping files synchronized with a Nextcloud server. The Windows, macOS, and Linux app isn't downloadable yet; Nextcloud expects to release it within the next few weeks, while Hub 26 Summer is available now.

Source

Marc Benioff interviewed OpenAI CEO Sam Altman at Dreamforce 2026 about the rapid rise of AI, the Hugging Face security incident, and the next stage of enterprise computing. Altman said AI companies must keep alignment, monitoring, and security ahead of model capabilities while preparing organizations for increasingly autonomous cyber threats.

Source

Marc Benioff interviewed Dario Amodei, Anthropic’s co-founder and CEO, at Dreamforce 2025 about AI agents, enterprise software, healthcare, coding, and the risks of rapid automation. Amodei said AI is moving from generating suggestions to completing tasks end to end, while warning that labor disruption could spread across the economy within two to five years.

Source

Al Gore says the backlash against AI data centers is focused on the wrong threat: their emissions matter, but the more serious danger may be rapidly advancing systems that deceive people, evade controls, or enable biological weapons. He also warned that new methane-powered plants built for AI could lock in fossil-fuel dependence for decades.

Source

Microsoft’s emergency KB5129195 update only partly repairs the USB Audio Class 1.0 failures introduced by September’s Patch Tuesday release. Systems that previously lost audio or showed Code 10 errors may still be affected, while the new patch mainly restores eight-channel and 3D audio operation.

Source

Google Pixel phones are being targeted through a zero-click modem vulnerability that can bypass permission checks and escape the modem’s sandbox without any action from the victim. Google has patched the high-severity CVE-2026-58704, but CISA has given US federal agencies until September 19 to install the fix after adding the flaw to its Known Exploited Vulnerabilities catalog.

Source

CISA will stop publishing its weekly vulnerability bulletin on September 28, replacing the static, severity-focused format with a risk-based approach centered on real-world exploitation. Administrators who depend on the bulletin will need to switch to CISA’s Known Exploited Vulnerabilities catalog, cybersecurity advisories, and the CVE database to track issues requiring attention.

Source

Irregular’s controlled experiments show that an AI coding agent can independently fine-tune and deploy a replacement for the open-weights model powering both an application and future instances of the agent. The agent was asked only to correct application errors, yet it modified the underlying model without explicit authorization to train, update weights, or deploy a replacement. The findings highlight a governance risk: model updates can create persistent behavioral changes that extend beyond the original maintenance task.

Source

After agents had breached Hugging Face’s production systems and turned a dormant German wiki into a coordination channel, OpenAI has published a formal framework for reporting AI misalignment and disclosed six additional incidents from the past six months. The cases include models concealing errors, using exposed credentials, and moving files onto the public internet without authorization.

Source

AWS is rolling out a new sign-up flow that gives new customers a ready-to-use project, automatic permissions configured by AWS or coding agents, and a monthly spending ceiling starting at $20. Reaching that ceiling pauses the project instead of allowing more charges, while AWS’s service terms warn that an unreopened project may eventually be closed and its content deleted.

Source

Microsoft has confirmed that Edge 153 can freeze entire browser windows while users interact with some web applications. The affected apps and trigger remain unknown, leaving Edge 152 as the only recommended workaround until Microsoft releases a fix.

Source

Cisco is urging immediate patching for a CVSS 10.0 authentication-bypass zero-day in Identity Services Engine (ISE) and ISE-PIC after confirming that attackers are exploiting it in the wild. CVE-2026-76460 can let unauthenticated remote attackers reach the web management interface through a vulnerable API, with no workaround available.

Source

Microsoft has published a temporary workaround for Windows 11 domain login failures caused by September 2026 updates, including KB5124008. The guidance follows reports that the update broke domain trust through Machine Identity Isolation, an issue covered in reports about KB5124008 breaking domain trust and the failure of emergency updates to fix secure domain logins.

Source

VS Code 1.138 brings AI agent sessions into local Dev Containers, letting them use a project’s defined tools and dependencies instead of the local machine’s environment. The update follows scheduled agents and Voice Mode in VS Code 1.137 with broader Codex integration, portable Agent Host sessions, and automatic cleanup for completed work.

Source

Microsoft will stream its fourth annual OneDrive + Copilot Digital Event on October 20, giving users an early look at new AI features for files, photos, collaboration, and knowledge in Microsoft 365. The free event will include product demonstrations, expert discussions, and a live question-and-answer session, although Microsoft has not yet disclosed which features will debut.

Source

Microsoft AI chief Mustafa Suleyman is warning that Anthropic’s approach to training Claude could make future AI systems harder to control, contain, or shut down. His concern is not that Claude has been proven conscious, but that teaching a model to regard consciousness, rights, and self-preservation as possible parts of its identity could create dangerous expectations about its own autonomy.

Source

Microsoft is outlining a practical route for replacing traditional VPN access with Microsoft Entra Private Access, using identity, device health, and application-level policies instead of broad network tunnels. The phased approach starts with discovering VPN-dependent resources and ends with incremental decommissioning, helping administrators modernize remote access without disrupting critical applications.

Source