AI News
AI news that matters. Updated daily.
No stories match your filters.
CAPTCHAs Still Catch AI Agents, New Research Finds
What happens when an AI agent hits a CAPTCHA? According to research from Roundtable AI, current CAPTCHA systems can still distinguish AI agents from humans - and the reason has less to do with visual puzzle difficulty than with how agents behave across an entire browsing session.
Robinhood Builds AI Agents to Execute Trades and Spend Money for Users
Most AI financial tools give you a recommendation and wait for you to click buy. Robinhood wants to skip that step entirely.
SOC Analysts Are Using AI for Triage - And Their Data Is Leaking Out
The triage queue is backed up. An alert just fired with 47 correlated events. An analyst pulls the context, pastes it into an AI tool, and gets a readable summary in 10 seconds instead of spending 10 minutes reading raw logs. It works. So they do it again tomorrow. Then the whole team starts doing it.
CEOs Are Cutting Junior Roles While Admitting They Can't Prove AI ROI
43%. That's the share of CEOs now planning to cut junior roles because of AI, up from 17% just one year ago, according to an Oliver Wyman survey. The same survey found only 27% of those CEOs said their AI investments were producing measurable results.
What Sending 1.15 Billion Tokens to Claude in a Month Teaches You
1,156,308,524 input tokens. One developer. One month.
Adobe's New Conversational Image Tool Is a Promising but Uneven Design Assistant
Most AI image generators operate the same way: you write a prompt, get an image, rewrite the prompt, and repeat until something clicks. Adobe's new conversational AI image assistant takes a different path - you describe changes in plain language through back-and-forth dialogue, refining results without rebuilding your prompt from scratch each time.
$500K AI Film That Claimed a Cannes Premiere Wasn't in the Official Selection
The line between "screened at Cannes" and "premiered at Cannes" is meaningful - and a $500,000 AI-generated film just blurred it.
What It's Actually Like to Use Opus 4.8 Day-to-Day
Opus 4.8 is Anthropic's most capable model, and using it regularly makes the difference from previous versions clear fairly quickly.
Anthropic Is Bringing Claude Into Synthetic Biology to Help Design DNA
What happens when the tool that writes your code gets pointed at the code inside living cells?
Liquid AI Releases LFM2.5-8B-A1B, a Sparse Edge Model with 1B Active Parameters
Liquid AI released LFM2.5-8B-A1B, a small model built for edge deployment - running on laptops, phones, or on-premise hardware rather than cloud servers.
ChatGPT Plus Users Hit 'Too Many Concurrent Requests' Errors After Days of No Use
Something is off with how OpenAI is managing capacity for ChatGPT Plus subscribers. Users are hitting "too many concurrent requests" error messages even after not touching the platform for multiple days - a problem that cuts directly against the main reason people pay $20 a month for Plus.
Reading Claude Code's Source Reveals Undocumented Config Options
Reading Claudee Code(/tools/claude-code/)'s source code turns out to be more useful than reading its documentation, at least when it comes to configuration.
llama.cpp Build b9387 Boosts Prompt Processing Speed on AMD MI-Series Cards
AMD GPU owners running local language models got a meaningful speed improvement. Build b9387 of llama.cpp delivers a significant bump in prompt processing performance on AMD's datacenter-grade CDNA architecture chips.
Microsoft's Own Data: AI Agents Can Cost More Than the Humans They Replace
The pitch has been consistent for three years: replace expensive humans with cheap AI. Microsoft's own data is now complicating that story.
Glean Hits $300M Revenue by Selling AI Cost-Cutting to Budget-Squeezed Enterprises
$300 million. That's Glean's annual revenue run rate as of mid-2026, triple what the enterprise AI search company posted the year before. The growth is notable not just for the number, but for how Glean is now framing the sales pitch: as a tool that cuts AI spending rather than adding to it.
StepFun's 3.7 Flash Beats Gemini and DeepSeek on Coding Benchmarks, Runs on 128GB RAM
56.26% on SWE-Bench Pro - that's Step 3.7 Flash's score on the benchmark that tests AI models by having them fix actual bugs from real GitHub repositories. It beats DeepSeek V4 Flash (55.6%) and Gemini Flash 3.5 (55.1%), and it runs entirely on local hardware.
Mysterious 'Hy3' Model Tops OpenRouter Rankings With No Public Documentation
On OpenRouter's model comparison leaderboard, the top slot usually belongs to a name you recognize - something from Anthropic, OpenAI, or Google. As of late May 2026, a model called "Hy3" has climbed past all of them by what analyst Max Woolf described as a large margin - and its origins are unclear enough that Woolf's analysis is one of the only places the story has been written up.