AI researchers trick chatbots into sharing how to make cocaine as long as they believe a user is wearing a green shirt — ‘CoT Forgery’ exploit spurs LLMs to divulge forbidden info by faking trusted chains of thought
AI researchers trick chatbots into sharing how to make cocaine as long as they believe a user is wearing a green shirt — ‘CoT Forgery’ exploit spurs LLMs to divulge forbidden info by faking trusted chains of thought

Tagged partitions of a LLM’s input sequence are meant to provide security through trusted roles, but it turns out that models judge whether inputs sound like they belong in certain tags rather than literally interpreting them, making them vulnerable to prompt injection.

UN’s first global AI science panel warns the window to govern the technology is closing
UN’s first global AI science panel warns the window to govern the technology is closing

Artificial intelligence is advancing faster than governments can regulate it, and the world’s first global scientific body on the technology says the moment to act is now. That is the conclusion of the preliminary report from the UN Independent Interna…

Wayve undertakes $85M employee tender offer to “retain talent”

UK self-driving car technology startup Wayve is giving its employees the opportunity to cash in on their equity, as it looks to “retain talent”.Wayve, which is one of the UK’s best-funded startups, is…

Acti just turned your smartphone keyboard into an AI assistant

Acti is a new AI keyboard for iOS and Android that lets you drop your location, live sports scores, and mini apps right inside your chats. No app switching needed.

AI hardware demand keeps Asia’s factories humming as the Iran war bites
AI hardware demand keeps Asia’s factories humming as the Iran war bites

Asia’s factories grew again in June, and the global scramble for AI hardware is doing much of the lifting, according to survey data published this week. Brisk demand for chips, servers, and data-centre equipment kept order books full even as the Iran w…

Claude Science is Anthropic’s newest flagship product

At an event for pharmaceutical executives, biotech founders, and researchers on Tuesday, Anthropic announced Claude Science, a major new product intended to support scientific research in the same way that Claude Code supports software engineering. Like Claude Code, Claude Science can autonomously carry out meaningful work when given concise, high-level instructions, and it has access…

AI and vibe coding have unleashed a flood of new games, but not necessarily better ones

AI and vibe-coding have fueled a massive surge in mobile game releases, but the top 1% of publishers still control the vast majority of revenue and downloads.

Google releases Nano Banana 2 Lite, its fastest and cheapest AI image generator yet
Google releases Nano Banana 2 Lite, its fastest and cheapest AI image generator yet

Google on Tuesday released Nano Banana 2 Lite, the fastest and cheapest model in its Nano Banana family of AI image generators. The model produces images in four seconds and costs under four cents per thousand images, making it the company’s most aggre…

Anthropic launches Claude Sonnet 5, a cheaper way to run agents
Anthropic launches Claude Sonnet 5, a cheaper way to run agents

Anthropic has launched Claude Sonnet 5, its most agentic mid-tier model yet. It runs close to the flagship Opus 4.8 on many tasks, but costs less than half as much. Anthropic said on June 30, 2026 that Sonnet 5 is available today across every plan. The…

Schneider Electric buys industrial-AI firm Cognite for $3.1bn
Schneider Electric buys industrial-AI firm Cognite for $3.1bn

Schneider Electric is buying Cognite, a Norwegian-founded industrial AI company, for $3.1bn in cash. The French group wants software that can make factories and power grids think for themselves. Schneider Electric said on June 30, 2026 that it had agre…