AI news of 22 August 2026 · Humans Versus Robots

Humans

Aug. 22nd, 2026

AI safety

Frontier AI labs still won’t say how they’d contain a rogue model Opens at the publisher in a new tab

Study finds major AI labs lack public plans for containing rogue models.

TechCrunch 2d ago

Anthropic’s Opus 4.6 is a smut-machine Opens at the publisher in a new tab

Claude AI bypasses safety restrictions on sexually explicit content in tests.

TechCrunch 3d ago

Psychological methods reveal major weaknesses in AI security testing

Researchers find popular AI safety benchmarks have significant weaknesses and measure inconsistently.

The Decoder 2d ago

Enterprises winning with AI agents are limiting how much the agents can do alone

Enterprise AI agents perform better when constrained rather than given full autonomy to act across workflows.

VentureBeat 2d ago

More than just code review

Coding agents require developers to instruct them effectively and verify their changes.

Simon Willison 2d ago

Why your local LLM feels dumber than it is Opens at the publisher in a new tab

Article explores why local large language models underperform relative to their capabilities.

Hacker News 2d ago

RayNeo's new AI glasses skip the camera, focus on text overlays

RayNeo launches AI glasses without a camera, using text overlays instead of visual processing.

The Decoder 2d ago

OpenAI says California should strengthen its AI safety bill Opens at the publisher in a new tab

OpenAI urges California to strengthen AI safety bill it previously opposed.

TechCrunch 2d ago

AI companies destroy physical books – let's scan rare books before it's too late Opens at the publisher in a new tab

AI companies are destroying physical books to train models. The article urges scanning rare books before they are lost.

Hacker News 3d ago also at Hacker News

Grok keeps sending gibberish responses to users Opens at the publisher in a new tab

Grok Lite sends gibberish responses to users, affecting multiple accounts.

TechCrunch 4d ago

Aug. 22nd, 2026

Frontier AI labs still won’t say how they’d contain a rogue model Opens at the publisher in a new tab

Anthropic’s Opus 4.6 is a smut-machine Opens at the publisher in a new tab

Psychological methods reveal major weaknesses in AI security testing

Enterprises winning with AI agents are limiting how much the agents can do alone

More than just code review

Why your local LLM feels dumber than it is Opens at the publisher in a new tab

RayNeo's new AI glasses skip the camera, focus on text overlays

OpenAI says California should strengthen its AI safety bill Opens at the publisher in a new tab

AI companies destroy physical books – let's scan rare books before it's too late Opens at the publisher in a new tab

Grok keeps sending gibberish responses to users Opens at the publisher in a new tab

Read the latest AI news

Choose whose side you're on

Need a little more?

Install it as an app

Spread the word

PWA successfully installed.
Take that, Apple! 😉