AI News
AI news that matters. Updated daily.
No stories match your filters.
GPT-Next Solves an 80-Year-Old Math Problem for Less Than $1,000
Under $1,000. That's the compute bill OpenAI's GPT-next reportedly ran up to resolve the Erdős unit distance problem, a combinatorics conjecture that stumped professional mathematicians for 80 years.
OpenRouter Raises $113M Series B at $1.3B Valuation as Multi-Model Usage Surges
In a year, OpenRouter's valuation has more than doubled to $1.3 billion. The company just closed a $113 million Series B led by CapitalG, Google's independent growth fund, and the numbers behind the raise tell you something real about how developers are actually building with AI right now.
Anthropic's Cybersecurity AI Mythos Shows Results But Stays Out of Public Reach
Anthropic updated the public status of Mythos, its AI model built to automatically find security vulnerabilities in software. The short version: the model is capable, and no one outside Anthropic is getting access to it yet.
Token Leaderboards: How One Company Gamified Claude Code Adoption
Token leaderboards are apparently a thing now.
Mistral AI Partners With Harvey to Break Into Legal Market
Harvey, the legal AI startup that has signed up dozens of Am Law 100 firms, is now running Mistral AI's models alongside its existing lineup. The partnership gives Mistral a direct route into one of the most lucrative enterprise verticals in AI - legal work - without having to build client relationships from scratch.
Universal Music Group and TikTok Renew Deal to Fight Unauthorized AI Music
Two years after Universal Music Group pulled its entire catalog from TikTok over a licensing dispute, the two companies have renewed their agreement - this time with explicit provisions targeting AI-generated music that uses artists' voices or works without permission.
Sundar Pichai on AI Search, Zero-Click Queries, and What's Left of the Open Web
Five years in, Sundar Pichai's annual post-Google I/O conversation with The Verge's Decoder has become the closest thing to a state-of-the-union for anyone whose business depends on Google Search. This year's recording lands at a moment when those stakes feel genuinely different.
AI Is Making HTML a Practical Replacement for PowerPoint on Client Deliverables
A year ago, asking an AI to generate a client-facing proposal as HTML instead of a Word doc or PowerPoint deck was an interesting experiment with an obvious catch: the output looked like a developer built a slide deck at midnight. Design was generic, PDF export was messy, and handing a client an HTML file felt unprofessional enough to undermine the whole thing.
AI Image Generators Worth Paying For: A Business Buyer's Breakdown
Spending a few thousand dollars a year on AI image generation is a real budget decision for a growing number of businesses. After testing most of the serious contenders, the honest answer is: the "best" generator depends almost entirely on whether you can afford to get sued.
China Restricts Overseas Travel for AI Researchers at Alibaba and DeepSeek
China is blocking AI researchers and engineers at Alibaba and DeepSeek from traveling overseas, according to reports surfacing this week. The restrictions apply specifically to staff working on AI projects, and the move is a direct attempt to stop US and European companies from poaching the people behind some of China's most competitive AI models.
AI Coding Agents Rewrote Software Development in 18 Months
The past 18 months broke something in software development, not in a damaging way, but in the way a door breaks open. AI coding agents went from writing boilerplate snippets to completing entire projects with minimal human direction, and Wired's account of the transformation positions Claudee Code(/tools/claude-code/) and OpenClaw as the tools that made this shift irreversible.
A Professional Fact-Checker's Verdict on AI Accuracy: More Wrong Than You'd Expect
What happens when someone whose entire job is catching errors puts AI through its paces? A professional fact-checker at Wired did exactly that, and the conclusion is worth sitting with: AI gets things wrong more often than most users realize.
How to Use AI Tools Like a Power User, Not a Beginner
There's a growing gap in most workplaces between people who use AI and people who work in AI. The first group opens ChatGPT, types a question, copies the answer, and moves on. The second group has restructured how they think, what they save, and how they scope projects - all around what AI can and can't reliably do.
Uber Burned Through Its 2026 AI Budget in 4 Months and Can't Show the Returns
Four months. That's how long it took Uber to exhaust its entire annual AI budget in 2026. By April, the company had already hit its ceiling - and according to Andrew Macdonald, Uber's president and chief operating officer, they don't have a clean answer for what they got in return.
Anthropic Launches CCA-F: A Certification Exam for Claude Architects
The Claude Certified Architect - Foundations (CCA-F) exam is now live - Anthropic's first formal credential for developers who design and build systems using their AI models.
Claude's "Honest Caveat" Habit Is Getting on Users' Nerves
Something has shifted in how Claude phrases its responses, and frequent users are noticing. The phrase "honest caveat" has become a common sentence opener in Claude's replies, appearing in lines like "Honest caveat: I can't verify this in real time" or "Honest caveat: this approach has tradeoffs." Users who work with the model daily are flagging it as a new behavioral tell.
Self-Hosting LLMs Isn't Cheaper. Here's the Actual Math.
The "local is cheaper" argument for running AI models on your own hardware gets repeated constantly in the self-hosting community. The math doesn't actually support it.
AI Proves 44 New Math Conjectures and Solves 9 Problems Unsolved for Decades
Nine novel math problems solved. Forty-four new mathematical conjectures proved. Some of these problems had resisted solution for 50 years.
Wiz Connects to Anthropic's API to Bring Claude Into Cloud Security Workflows
Cloud security platform Wiz has integrated with Anthropic's API, giving security teams access to Claude from inside their existing Wiz dashboards. The integration is detailed in a post on the Wiz blog.
IBM Quantum Computer Helps Train AI That Outperforms Its Classical Baseline
Researchers used an IBM quantum computer to fine-tune an AI language model, and the result correctly answered questions the original, classically-trained version got wrong.
Users Report Claude Inserting Unexplained Injection Prompts Mid-Conversation
Some Claude users have noticed unexpected text appearing at the end of the AI's responses - text that reads like a system-level instruction rather than a natural reply. What they're seeing is what security researchers call a "prompt injection": an instruction embedded in text that can redirect an AI's behavior. The strange part is that the injections appear to be coming from Claude itself, not from any external content trying to manipulate it.