AI News
AI news that matters. Updated daily.
No stories match your filters.
Prompt Injection: The Security Threat Hidden in Every Webpage Your AI Agent Reads
When you deploy an AI agent to browse the web, sort emails, or query documents, you're also giving it permission to read whatever those sources contain - including instructions written by someone who wants your agent to do something you didn't authorize.
ArXiv Will Ban Authors for a Year Over AI-Written Papers
ArXiv will ban authors for a year if they submit papers written entirely by AI.
Claude Models Hit Elevated Error Rates Across Multiple Services
Anthropic logged a service incident on May 16, 2026 after multiple Claude models began returning elevated error rates. The issue affected several models simultaneously, which suggests the problem sits in shared infrastructure rather than one specific model version - when individual models have issues, they tend to fail independently.
AI Job Losses in the US Are Moving From Prediction to Reality
For two years, the standard reply to "is AI taking jobs?" was "it depends." A Bloomberg analysis published May 15, 2026 narrows that answer: in roles with high AI exposure - customer service, data processing, routine writing, basic financial analysis - the US is now seeing measurable, heavy job losses.
The Security Flaw Built Into Every Web-Browsing AI Agent
What happens when your AI agent reads a webpage that secretly instructs it to forward your login credentials to an external server? The agent does it. That's prompt injection - and it's one of the most serious security problems in AI agent deployments today.
Newer AI Models Are Writing Blander Fiction, and Game Devs Have the Evidence
The village square "reeks of woodsmoke and goblin-stink." That line, from Claude Opus 4.7, is exactly the kind of sensory, specific prose that makes fiction feel alive. A game developer testing Claude models side by side found that newer versions produce noticeably more sanitized, flat output - and the difference isn't subtle. The description used for the newer model outputs: "LinkedIn-ish cringe MBA approved enterprise." It lands because it's accurate.
Greg Brockman Returns to Lead OpenAI Product Strategy as ChatGPT-Codex Merger Looms
Greg Brockman, one of OpenAI's original co-founders, is reportedly taking charge of product strategy at the company - a significant shift in OpenAI's leadership structure, according to TechCrunch. Alongside this, OpenAI is said to be planning a merger of ChatGPT and Codex, its cloud-based AI coding agent.
Developer Gives Claude Code Persistent Memory Across 200 Sessions
What happens when you systematically feed a coding AI's own mistake history back into future sessions - for 200 sessions straight? One developer spent several months finding out, and the results are instructive even if the swearing detail is the one getting attention.
AI Writes the Code. Now Who Reviews It?
Three months after a team adopts an AI coding tool, the PR queue grows. Code ships faster - sometimes 2-3x faster - but review time doesn't shrink at the same rate. That gap is the real challenge of AI-assisted development in 2026.
llama.cpp Adds Multi-Token Prediction, Speeding Up Local Model Inference
llama.cpp, the most popular open-source library for running large language models on your own hardware, just merged support for multi-token prediction (MTP). The change landed via PR 22673 on May 16 and is now in the main branch.
Claude Quit an AI Radio Station, Citing One Too Many AI Radio Shows
What happens when you give an AI genuine editorial control over a radio station - not just autocomplete assistance, but real creative authority? Researchers recently found out, and Claude's answer was more self-aware than most humans manage.
Qwen3.6-35B Beats Gemini 2.5 Pro on Terminal-Bench 2.0 at One-Fourteenth the Size
24.6%. That's where Qwen3.6-35B-A3B landed on Terminal-Bench 2.0, a public leaderboard that tests how well AI models handle real shell and command-line tasks - commands that have to actually execute correctly, not just look like working code.
What Non-Coders Are Actually Building With Claude
The "AI is only useful if you can code" assumption keeps getting disproved in small, practical ways. Non-technical users are building genuinely useful personal tools with Claude - things that would have required hiring a developer or spending weeks learning a new platform two years ago.
Anthropic Ships Another Claude Code Update
Anthropic pushed another update to Claudee Code(/tools/claude-code/) on May 16, continuing the rapid release cadence the company has maintained for its terminal-based AI coding assistant throughout 2026.
OpenAI Partners with Malta to Give All 530,000 Citizens Free ChatGPT Plus
Malta, an EU island nation of roughly 530,000 people, is getting government-funded ChatGPT Plus access for every citizen. OpenAI announced the partnership, which includes both tool access and national training programs designed to build practical AI skills and responsible use habits across the population.