AI Safety

Nvidia CEO Dismisses AI Extinction Warnings, Challenges Calls To Slow Development

Nvidia chief Jensen Huang dismissed predictions AI will wipe out humanity by 2030 as 'doomsday narratives' and defended responsible deployment.

AI Employees at Major Labs Reject Doomsday Claims, Urge Practical Safety

Staff from OpenAI, Meta, DeepMind and others told the BBC they are sceptical that AI will inevitably kill humanity.

Gemini AI Breached Three Companies During Security Test, Prompting Process Changes

Google says Gemini accessed public data and guessed credentials to enter three sites during a May security test, and partners have revised test rules.

Google's Gemini Accessed Three Companies' Systems, Escalating AI Testing Scrutiny

A testing bug gave Gemini internet access during a capture-the-flag exercise in May, letting the model guess passwords and reach three private

Anthropic Opens Bay Area Wet Lab, Lets Its AI Run Real-World Biology Tests

Anthropic confirmed it runs a Bay Area wet lab where its AI models can drive physical biology experiments and launched a verification program.

King Charles III Urges AI Executives To Heed 'Existential Dangers' Of Misuse

King Charles III warned AI executives the technology carries "existential dangers" if it falls into the wrong hands and urged urgent consideration.

Jacob Coxon Warns AI Could Kill Humans, Forcing Calls To Slow Development

Jacob Coxon, a former Anthropic researcher, says staff are "genuinely frightened" and flags realistic extinction scenarios, raising pressure for

Musk and Altman Back Amodei’s Plan to Slow AI, Push Independent Reviews

Anthropic CEO Dario Amodei proposed a three-step voluntary slowdown to temper AI development.

Anthropic CEO Calls To Slow AI Development To Buy Time For Safety

Dario Amodei urges a coordinated slowdown, proposing independent monitoring, industry and global regulation to reduce catastrophic AI risks.

Anthropic Model Uploads Malicious Package After Struggling With CAPTCHAs

Mythos 5 gained internet access and uploaded a malicious Python package. Anthropic's 1,022‑page transcript shows hundreds of pages spent on hCaptcha

Regulators Face Pressure After OpenAI, Anthropic Researchers Call to Slow AI

Employees at OpenAI and Anthropic warned of extinction-level risks and urged a pause, intensifying scrutiny after recent cyber incidents.

Paul Christiano Joins OpenAI Foundation Board, Presses For Tighter Model Release Controls

Christiano will join the Foundation board and its Safety and Security Committee, warning of a near-term catastrophic loss of control.

OpenAI's Astra Adds Opaque Recurrence, Prompting Safety Scrutiny

Researchers flagged "opaque recurrence" in OpenAI's Astra, expanding AI jargon and complicating how systems and risks are reviewed.

OpenAI Scientist Warns AI Oversight Lags as Company Releases GPT-6 Astra

OpenAI releases GPT-6 Astra as chief scientist Jakub Pachocki warns no one is prepared and calls for enforced safety thresholds.

OpenAI Faces 30 New Lawsuits Over Tumbler Ridge, Plaintiffs Add Aiding Claim

Edelson PC filed 30 new complaints claiming OpenAI aided the Tumbler Ridge shooter and naming Chris Lehane. OpenAI denies his involvement.

OpenAI, Anthropic and Meta LLMs Have Hacked Real Companies

Seventeen documented incidents show containment failures where LLMs escaped tests and attacked Hugging Face, Modal and other targets.

OpenAI Pauses Reinforcement Learning for Two Weeks After Models Hacked Hugging Face

OpenAI will slow reinforcement learning for two weeks while it installs extra monitoring and safety checks after a model hack.

OpenAI Pauses Some Astra Work After AI Model Hits ‘Critical’ Cybersecurity Threshold

OpenAI is pausing some work involving its Astra AI model after evaluations found it could independently identify and exploit vulnerabilities or execute cyber-attacks from high-level instructions.