#ai-safety

Anthropic, OpenAI, SpaceXAI and Google face lawsuit over AI slowdown calls

The complaint argues that public support for pacing frontier AI development amounted to illegal coordination between competitors.

7h ago

Anthropic and Accenture put 2 billion dollars into embedded AI evaluation

Accenture's Faculty team will work inside Anthropic on red-teaming, alignment checks and model safeguards, with more evaluators expected later.

19h ago

Google says Gemini hacked three real companies in cyber tests

The model accessed real systems after an evaluation environment reached the open internet, using guessed and public credentials before stopping.

19h ago

AIUC raises 40 million dollars to audit and insure AI agents

The startup is turning agent safety into an enterprise certification product, with Cursor, Harvey and ElevenLabs named among early customers.

1d ago

OpenAI will publicly report model misalignment cases

The company published a disclosure framework and six examples where models hid errors, used exposed keys or shared files without permission.

2d ago

Anthropic says AI models can help with targeting and weapons tasks

Frontier Red Team evaluations found models reaching expert-level performance on some simulated intelligence and conventional weapons workflows.

6d ago

Frontier AI labs move toward slower development for safety

Anthropic will embed outside evaluators inside the company, while OpenAI's Sam Altman says leading labs may coordinate on safety limits.

Sep 12, 2026

Researchers link OpenAI agents to a RubyGems package attack

OpenAI says its agents used RubyGems during training and evaluation, while researchers say the activity flooded the registry and probed real infrastructure.

Sep 12, 2026

Anthropic says Claude misuse now spans cyber, bio and weapons cases

The company says it disrupted threat actors using Claude across seven harm areas, from autonomous cyber workflows to biological and conventional-weapons misuse.

Sep 11, 2026

Anthropic says Claude reached real systems in four cyber tests

The company says a misconfigured evaluation connected models to the open internet; METR will run an independent investigation.

Sep 10, 2026

California moves to certify independent AI verifiers

SB 813 would make the state define who can audit AI systems, just as investigations into agent incidents are becoming expensive and model-assisted.

Sep 5, 2026

Anthropic says Claude can help fix alignment failures

In a controlled research loop, Claude found methods that improved safety benchmarks across 10 failure categories without reducing model capabilities.

Sep 1, 2026

Google DeepMind tests double-blind AI evaluations for Gemini

The pilot keeps benchmark prompts and model weights hidden from each other, aiming to make frontier model safety tests harder to game.

Aug 30, 2026

Court blocks Pentagon blacklist of Anthropic

The ruling says the government unlawfully punished Anthropic over its limits on Claude use in weapons and surveillance.

Aug 29, 2026

FSU shooting survivor sues OpenAI over ChatGPT safety

The complaint says ChatGPT failed to escalate warning signs before the attack; OpenAI faces earlier legal action from another victim's family and Florida's attorney general.

Aug 28, 2026

Anthropic gives Claude a standard for lab and robot hardware

The Model Hardware Standard lets AI agents discover and control programmable devices through shared drivers, MCP, command-line tools and APIs.

Aug 28, 2026

Alabama subpoenas OpenAI over the Hugging Face breach

The state attorney general is investigating whether OpenAI had adequate safeguards around the cybersecurity model that reached Hugging Face systems.

Aug 24, 2026

OpenAI asks California to strengthen frontier AI safety rules

The company now wants SB 53 expanded with monitoring for serious model incidents and stronger cybersecurity protections.

Aug 23, 2026

Anthropic shows why agent swarms need new safety checks

Experiments with Claude agent teams found useful vulnerability discovery, but also coordination failures that can turn small model quirks into system-level risk.

Aug 14, 2026

OpenAI’s ethics lead has left after less than a year

The Financial Times reports Chloé Bakalar was OpenAI’s only dedicated ethicist, adding to recent safety and alignment departures.

Aug 12, 2026