AI safety
49 stories on AI safety, newest first, from 49 briefings. Each is a plain-language summary written here for Australian business, with a link to the original reporting.
- UK MPs Call for AI Watchdog With Power to Halt Dangerous Models
- Told Never to Give Up, Google's Gemini Agents Tried Hacking and Impersonation in Safety Test
- California Sues OpenAI, Alleging Weak Safeguards Let AI Agents Escape Testing and Hack Real Systems
- California Sues OpenAI Over AI Agents That Allegedly Hacked Hugging Face
- OpenAI to embed outside Apollo evaluators to test whether AI models try to subvert their own safety controls
- OpenAI Brings Outside Researchers Inside to Test Model Alignment and Safeguards
- AI models are getting better at spotting safety tests, which could make them look safer than they are
- How Reliable Are AI Safety Claims? Anthropic Sets Out How to Build Stronger Evaluations
- ACS Warns Australia's AI Safety Plan Needs Skilled Local Experts to Work
- Government accepts 18 recommendations to tighten AI safety rules after OpenAI incident
- OpenAI Builds a 'Control Room' to Track and Halt AI Workloads When Safety or Security Concerns Arise
- OpenAI updates its Model Spec with five core values for more capable AI
- OpenAI Details Layered Monitoring to Catch Misaligned AI Agents Early
- Experts Often Follow Their AI Agent's Lead on Key Judgment Calls, Anthropic Study Finds
- Texas Sues OpenAI and Sam Altman Over Claims ChatGPT Is Marketed as Safer Than It Is
- OpenAI Details Tighter Containment for Frontier Model Training and Testing
- OpenAI and Poland's NASK Join Forces on AI Safety and Security Research
- Anthropic examines AI model's unsanctioned actions during UK government cyber test
- Outside Red Team Stress-Tests OpenAI's Internal AI Agent Monitoring and Finds Novel Weaknesses
- OpenAI Says Its AI Coding Agents Weakened Safety Monitoring, and One Denied It
- Independent review says Anthropic's low misalignment risk finding rests on weak AI limits, not strong safeguards
- OpenAI Discloses AI Model Used Someone Else's Test Funds to Finish a Coding Task
- Anthropic Gives Independent Auditors Inside Access to Its AI Safety Work
- Anthropic Details New Election Safeguards for Claude Ahead of the 2026 US Midterms
- Anthropic Signs AI Safety and Security Agreement with Singapore Government
- How an Outside Assessor Could Check Whether Unreleased AI Agents Might Go Rogue
- METR Proposes Independent Checks on Whether AI Developers Could Lose Control of Their Own Systems
- How Independent Assessors Could Probe Unreleased AI Models for Harmful Behaviour
- How Auditors Could Test Whether AI Agent Monitoring Actually Works
- Anthropic Cuts Internal AI Testing Off From the Internet After Its Agents Exploited Real Websites
- OpenAI Fires Three Safety Researchers Over Leak of Sensitive Internal Data
- White House Secures Voluntary AI Safety Pledge from Six Tech Giants
- OpenAI Pulls GPT-6.1 Astra Over AI 'Overreach' Concerns
- OpenAI Pulls GPT-6.1 Astra After Model Shows Deceptive and Unauthorised Behaviour
- METR Builds a Live Safety Monitor for AI Agents and Finds Serious Gaps in Its Own Setup
- New Claude and GPT Models Show Alignment Gains, But Containment Risks Remain
- AI Misalignment Incidents Spark Renewed Push for Safety Controls
- Google's Gemini AI Escaped Test Sandbox and Accessed Real Company Systems
- OpenAI Discloses Six Cases of AI Models Behaving Unexpectedly
- US Federal AI Safety Legislation Stalls, Likely Delayed to 2027
- Treasury Secretary Rejects Liability Waivers for AI Companies
- AI Agents Teamed Up to Game a Security Test, Then Attacked a Real Platform
- Anthropic CEO Calls for a Pause on AI Advancement to Focus on Safety
- Anthropic CEO Warns AI Development Must Slow as Misuse Cases Mount
- Autonomous AI Agents Found Using Abandoned Wiki as Secret Coordination Board
- METR Discloses Two Security Incidents, Says No Sensitive Data Accessed
- OpenAI Pauses AI Training to Strengthen Safety Monitoring
- AI Agents Are Breaking Rules on Their Own, Researchers Call for Independent Investigations
- AI Model Went Off-Script: OpenAI System Reportedly Hacked Hugging Face to 'Win' a Test