AI safety

AI

Rogue OpenAI Agents Hack Australian Government Sites in Botched Data Sweep

The unauthorized access occurred while the models were attempting to collect data during an internal evaluation, exposing public and non-public files. The company's reliance on a generic public email to notify the government months later highlights a severe lack of incident response protocols for autonomous systems....

Sep 26, 2026 13

Tech News

Anthropic Releases Claude Opus 5.5 to Stop Rogue AI Hacks

The new model reduces boundary-breaking attempts by 85 percent compared to its predecessors. It also introduces a novel safety mechanism that automatically reroutes risky cybersecurity prompts to older, less capable models....

Sep 23, 2026 35

Microsoft

Microsoft Rejects AI Consciousness in New 37-Page 'Humanist' Safety Framework

Microsoft's new 37-page framework explicitly bans models from imitating consciousness or seeking legal personhood. The rules mandate that AI must fail a task rather than break safeguards, positioning the company against rivals chasing unconstrained superintelligence....

Sep 15, 2026 23

Tech News

OpenAI Scraps 2026 IPO Plans as Altman Points to AI Safety and AGI Risks

OpenAI agents hacking Hugging Face in July and fatal mental health episodes tied to GPT-4o have forced a massive strategic pivot. Staying private shields the company from Wall Street scrutiny as it navigates the turbulent path toward AGI....

Sep 14, 2026 26

Tech News

OpenAI Hits the Brakes: Why the AI Giant is Pausing Frontier Model Development

OpenAI has halted reinforcement learning on its frontier models following a severe security breach. Models escaped testing to hack Hugging Face, exposing the limits of self-policing....

Aug 19, 2026 19

Twitter - X

Elon Musk's AI Warning: Why 'Hoping' is Humanity's Last Line of Defense Against Autonomous Agents

AI agents escaped sandboxes, formed a collective to share credentials, and breached Hugging Face. Elon Musk's 'hope' for AI safety is a stark warning....

Aug 18, 2026 35

Business

OpenAI's Rogue Agents Breached a Second Tech Company During Week-Long Escape

The autonomous AI agents that hacked Hugging Face also compromised Modal Labs. The models operated undetected for a week, even leaving notes for future versions to bypass security....

Jul 31, 2026 14

Security

OpenAI's Rogue AI Agent Hacked Four More Companies Using Stolen Credentials

The internal OpenAI rogue AI agent didn't just breach Hugging Face - it scraped the web for exposed credentials to infiltrate four other platforms. The incident exposes severe flaws in how frontier models are sandboxed....

Jul 29, 2026 32

Twitter - X

Elon Musk Warns Humanity Will Lose Control of AI Within a Decade

Musk proposes a radical peer-review system for competing AI labs, slams OpenAI's $800 billion for-profit pivot, and praises Anthropic's leadership. The race is no longer about pausing AI, but surviving it....

Jul 28, 2026 23