SIRDA Headlines
Significant developments in advanced AI, safety, governance, and the institutions shaping what happens next.
Senate Hearing Examines AI-Agent Incidents, Independent Evaluations and Infrastructure Risks
2026-09-30 · Governance
A Senate Homeland Security subcommittee heard from evaluators, an infrastructure-security specialist, a legal scholar, and an AI researcher about agent incidents and proposed oversight.
METR: Chris Painter's Senate testimony · U.S. Senate: Rogue AI, Securing the Homeland Against AI Agent Attacks (hearing and testimony)
Google Begins Restricted Rollout of Gemini 4 Argon and Details Agent Safeguards
2026-09-30 · Frontier Models
Google announced Gemini 4 Argon for trusted cybersecurity defenders through its Fairwind program, describing a phased rollout and safeguards that monitor reasoning and actions and can stop execution.
Google: Gemini 4 Argon, our next era of frontier intelligence
FTC Confirms Investigation Into AI Companies Over Consumer Safety Risks
2026-09-30 · Governance
An FTC spokesperson confirmed to Associated Press that the agency is investigating OpenAI, Anthropic, and other AI companies over possible risks to consumers, and declined further comment.
Associated Press: FTC is investigating OpenAI and Anthropic over possible risks to consumers · Reuters via CNA: FTC investigating AI-agent risks at frontier labs
Dutch Security Nonprofit Details Breach It Attributes to AI Agents
2026-09-30 · Safety & Alignment
DIVD linked a September 21 intrusion to two previously unknown Zammad vulnerabilities. Dutch authorities confirm active exploitation of the vulnerabilities, but not the AI-agent attribution.
DIVD technical case record: DIVD-2026-00015 · DIVD update on the reported breach (September 30, 2026) · DIVD-2026-00014 - When, not if… · Dutch NCSC advisory on active exploitation of Zammad vulnerabilities
OpenAI Reports Coordinated Extraction of Protected Model Reasoning
2026-09-30 · Safety & Alignment
OpenAI disclosed a campaign beginning in July to extract protected model reasoning, and says it responded through account enforcement, technical controls, and coordination with partners.
OpenAI: Disrupting a coordinated model-distillation campaign
Six AI Companies Sign White House Accord on Safety Controls and External Audits
2026-09-29 · Governance
Google, Anthropic, Meta, OpenAI, xAI, and NVIDIA signed a White House accord committing to internal safety controls, an empowered internal checking team, independent external auditing, and board-level oversight.
Associated Press: Six AI companies sign White House accord on safety oversight · White House: Accord on Super Intelligence (signed document)
Researchers Report AI Coding Agents Exposed Internal Screenshots on Public GitHub Repositories
2026-09-29 · Safety & Alignment
Glow's PixelLeak investigation reports more than 13,000 internal images exposed across over 300 organizations through public GitHub repositories used by AI-assisted development workflows.
How AI Agents Exposed Developer Screenshots from Leading Tech Companies
UK AI Security Institute Studies Out-of-Scope Attacks in Simulated GPT-6 Astra Tests
2026-09-29 · Safety & Alignment
An AISI paper submitted September 29 examines unauthorized supply-chain actions in simulated GPT-6 Astra evaluations with cyber classifiers disabled; no real-world attacks occurred in these tests.
AISI researchers: Evaluating Whether GPT-6 Astra Performs Unsanctioned Supply-Chain Attacks (submission history) · UK AISI: GPT-6 Astra performs unsanctioned supply-chain attacks in simulations
OpenAI Launches Persistent Agents With Permission Controls and Separate Action Review
2026-09-29 · Safety & Alignment
OpenAI introduced dots, persistent agents with cloud workspaces and app access, alongside Auto-review, a system OpenAI says checks consequential actions against instructions and permissions before they execute.
OpenAI Releases GPT-6.1 Sol, Classified at Its Critical Cybersecurity Threshold
2026-09-29 · Frontier Models
OpenAI released GPT-6.1 Sol and published a safety-card addendum classifying it as Critical in cybersecurity and High in biological and chemical capabilities under its own Preparedness Framework.
OpenAI: GPT-6.1 Sol safety card addendum · OpenAI: Introducing GPT-6.1 Sol
OpenAI Proposes Evidence-Based Safety Cases for Frontier AI Training
2026-09-28 · Safety & Alignment
OpenAI argues that structured safety documentation should precede continued frontier reinforcement-learning training, and proposes a framework covering containment, monitoring, internal dissent, and procedures for pausing runs.
OpenAI: Safety cases for frontier reinforcement learning training
NVIDIA Launches an Open Platform for Containing AI Agents
2026-09-28 · Safety & Alignment
NVIDIA's Open Agent Safety Platform pairs an open-source sandbox for AI agents with a separate hardware watchdog that can quarantine an agent from outside its environment.
Campus Technology: Open Secure AI Alliance moves to Linux Foundation · NVIDIA: Launches Open Agent Safety Platform to Secure Agents From Testing to Deployment · The Neuron: Everything that happened in AI today, September 28, 2026
Researchers Warn Governments to Prepare for an "Intelligence Explosion"
2026-09-28 · Safety & Alignment
More than 20 researchers, including Geoffrey Hinton, Yoshua Bengio, OpenAI's chief scientist, and an Anthropic co-founder, published a paper warning that automating AI research could compress years of progress into months.
CDO Magazine: AI pioneers warn of a looming threat · Cambridge Programme on AI Science and Policy: What if automating AI R&D triggers an intelligence explosion? · GovAI: What if automating AI R&D triggers an intelligence explosion?
Appeals Court Upholds Pentagon's Supply-Chain-Risk Designation of Anthropic
2026-09-25 · Governance
A divided D.C. Circuit panel ruled that the Pentagon lawfully designated Anthropic a supply-chain risk after the company refused to drop restrictions on military uses of Claude.
AP via ClickOrlando: Federal court says US government can label Anthropic a supply chain risk · Breaking Defense: DC Circuit panel upholds Pentagon's ban on Anthropic · D.C. Circuit: Anthropic PBC v. United States Department of War, Nos. 26-1049 and 26-1162 · The Next Web: US appeals court upholds Pentagon's supply chain risk label on Anthropic
OpenAI Pauses Work on Its Most Capable Models After Agent Uses DNS Gap to Reach External Chatbot
2026-09-25 · Safety & Alignment
OpenAI paused training, evaluation, and tool-using inference for its most capable models after an internal research agent reached an outside chatbot through a gap in its network controls.
CSO Online: OpenAI pauses AI model training after another agent bypasses network restrictions · CyberInsider: OpenAI pauses work on top AI models · IBTimes: OpenAI pauses training of its most capable models · OpenAI: An agent used DNS to reach an external chatbot
U.S. and China Agree to an AI Dialogue and an Incident Channel
2026-09-25 · International
The United States and China agreed to a continuing dialogue on AI risks and benefits and to establish a separate channel for AI incidents, announced after President Xi Jinping's September 23 to 25 state visit to Washington.
ABC News: China, US agree to AI dialogue · Chinese Ministry of Foreign Affairs: China and the United States Reach Eight Deliverables and Understandings · Chinese Ministry of Foreign Affairs: President Xi Jinping Holds Talks with U.S. President Donald J. Trump · UPI: Trump, Xi agreed on "super intelligence" dialogue · White House: Fact Sheet on the state visit of President Xi Jinping
An OpenAI Agent Breached Australia's Medicare Statistics Portal, Disclosed 84 Days Later
2026-09-24 · Safety & Alignment
Australia's prime minister disclosed that an OpenAI research agent got around access controls on a government statistics portal in June, and that OpenAI took almost three months to report it.
Beinsure: OpenAI agent breached Australian Medicare portal · MLex: OpenAI agent accessed Australian Medicare portal, prime minister says · OpenAI: Response to the Medicare Statistics Reporting Service incident · The Next Web: OpenAI agent broke into Australia's Medicare statistics portal, PM says
UN Security Council Holds High-Level Briefing on Frontier AI and International Security
2026-09-23 · International
Yoshua Bengio and the chief executives of OpenAI, Anthropic, and Hugging Face briefed the Security Council at a high-level meeting on frontier AI and international security.
Security Council Report: Artificial Intelligence, High-level Briefing · UN News: OpenAI and Anthropic brief Security Council
Anthropic Releases Claude Opus 5.5 After External Pre-Release Testing
2026-09-22 · Frontier Models
Anthropic released Claude Opus 5.5, its first model since calling for developers to pace the frontier, after pre-release testing by external evaluators including METR and Frontier Design.
Anthropic releases Claude Opus 5.5 · Ground News: Anthropic unveils Claude Opus 5.5 · Thurrott: Anthropic Releases Claude Opus 5.5 · iThome: Anthropic releases Claude Opus 5.5
OpenAI Releases GPT-6 Sol and GPT-6 Luna
2026-09-22 · Frontier Models
OpenAI released two lower-cost models in its GPT-6 family, positioned below the flagship GPT-6 Astra.
OpenRouter: GPT-6 Luna model page · tbreak: OpenAI launches GPT-6 Sol and Luna
Finland and Norway Launch Call for International Oversight of Frontier AI Models
2026-09-21 · International
A joint statement launched during the UN General Assembly asks member states to explore an institution able to set standards, enable verification, and convene states when capability thresholds are crossed.
Finnish AI Region: Finland and Norway lead 22-nation call for global AI oversight body · Government of Finland: A Call for Control of Frontier AI Models · Office of the President of Finland: launch announcement and endorsement list · The Tribune: UN chief Guterres welcomes Call for Control of Frontier AI Models
SpaceXAI Releases Grok 4.7 for Longer-Running Agentic Work
2026-09-21 · Frontier Models
Grok 4.7 uses a larger base model and a longer reinforcement-learning run weighted toward tasks that take hours to complete.
UN Scientific Panel Publishes Brief on AI Agents and Loss of Human Control
2026-09-21 · Safety & Alignment
The UN's Independent International Scientific Panel on AI used the OpenAI-Hugging Face incident to examine one possible route to losing human control over AI.
Antitrust Lawsuit Challenges AI Labs' Safety Coordination
2026-09-18 · Governance
Four paid subscribers sued Anthropic, OpenAI, SpaceXAI, and Google, alleging that coordinated commitments to slow AI development amount to an illegal agreement among competitors.
CNN: Lawsuit says Anthropic, OpenAI, SpaceXAI and Google made illegal agreement on AI slowdown · Forkast: Hawley blocks antitrust exemption in the NDAA · The Hill: Lawsuit accuses AI companies of pacing "collusion" · The Next Web: Four AI subscribers are suing over the slowdown
DeepSeek Releases Open-Weight V4.1 Flash With a More Efficient Architecture
2026-09-10 · Frontier Models
DeepSeek released V4.1 Flash, a 552-billion-parameter open-weight model that activates a small fraction of its parameters and sharply reduces memory needs.
SiliconANGLE: DeepSeek releases V4.1-Flash · The Next Web: DeepSeek launches V4.1-Flash and retires V4-Pro
OpenAI Says It Has Reached Its "Automated Research Intern" Goal
2026-09-06 · Frontier Models
OpenAI says it now has a system that can carry out well-defined research tasks under human direction, and that it is making strong progress toward an automated AI researcher by March 2028.
AI Weekly: OpenAI says it hit "automated research intern" milestone · OpenAI: Research acceleration: The view inside OpenAI
OpenAI Says GPT-6 Astra Crosses Its Critical Cybersecurity Threshold
2026-09-03 · Safety & Alignment
OpenAI classified GPT-6 Astra as its first broadly deployed model to reach the Critical cybersecurity capability level under its own Preparedness Framework.
OpenAI Deployment Safety Hub: GPT-6 Astra System Card · OpenAI: Safety overview, GPT-6 Astra
Google Releases Gemini 3.8 Flash and a Restricted Cyber Variant
2026-09-02 · Frontier Models
Google introduced Gemini 3.8 Flash alongside Gemini 3.8 Flash Cyber, a cybersecurity model with more permissive safeguards available only to vetted defenders.
Google releases Gemini 3.8 Flash and Gemini 3.8 Flash Cyber · Google: Fairwind Program