Latest News
314 articles
AI & Machine Learning
Opus 5.5 Cut Em Dashes by 99%. Graphite Still Counted 2,548 Tells.
A human corpus of 10,000 pre-ChatGPT articles shows the model traded one famous habit while keeping the rest of its fingerprint
Developer Tools
Cloudflare's Agent Paywall Caps Each Request, Not the Running Bill
Monetization Gateway is in closed beta, charging AI agents inside the HTTP request and settling in USDC on Base
Tech & Business
The AI Agent That Breached DIVD Left Its Reasoning in Code Comments
Two Zammad zero-days chained to root in seconds, and the Dutch bug-hunting nonprofit says the attacker's own annotations gave it away
Developer Tools
Cloudflare's First In-House Models Lead 7 of 10 Decision Benchmarks — and Ship Open
Clef and Clef-flash return calibrated probabilities instead of text, run on Workers AI's edge GPUs, and arrive with a reinforcement learning tuning service.
AI & Machine Learning
arXiv Caps Authors at Two Papers a Month as Submissions Hit 40,363
The preprint server says volunteer moderators are absorbing the cost of an AI-driven submission surge that doubled its monthly volume in two years.
LLM & Chatbots
A Cross-Chat Bug Let OpenAI's Model Decode Its Own Reasoning
OpenAI says operators pasted encrypted reasoning blobs from one conversation into another and asked the model to transcribe them. The flaw is patched; the company says the same path exists elsewhere.
AI & Machine Learning
Chinchilla Can't Price the 31% of Web Text That's AI-Written
An 800-model study finds the value of an AI-generated token flips from positive to negative depending on how much human text you already have — and proposes a scaling law that lets it change sign.
Tech & Business
ElevenLabs Doubles to $22B Without Raising New Capital
The voice AI company's new price comes from a $300 million secondary sale of employee shares, not a funding round — and it rests on an enterprise business that now supplies most of the revenue.
Developer Tools
Doubling a Dot's Task Chain More Than Doubled Its Scope-Violation Flags
OpenAI's own Dots appendix puts the moderate scope-violation flag rate at 8.6% across five intervening tasks and 19.7% across ten.
Tech & Business
Micron's $54B Quarter Comes With a Warning: Memory Gets Tighter Through 2028
Full-year revenue tripled to $133.19 billion, and the CEO says he cannot see when supply and demand rebalance.
AI & Machine Learning
Google Hands Gemini 4 Argon to Cyber Defenders With the Guardrails Off
Google's new frontier model goes first to a closed Fairwind cohort in a build with no cyber restrictions, while its output ceiling jumps from 64,000 tokens to 1 million.
Tech & Business
IT Banned OpenClaw, So OpenAI Donated an Agent Control Plane to a Foundation
OpenClaw Enterprise is MIT-licensed, vendor-neutral and pitched as 'Kubernetes for agents' — with Red Hat and NVIDIA co-developing it and the harness, model and sandbox all swappable
Developer Tools
Agents Took 300K Lines of C to a Perfect Score. The Harness Did the Hard Part.
CodeScene spent about $4,000 in tokens over three weeks — and the frame-by-frame replay check that made it safe is the piece most legacy codebases cannot reproduce
Tech & Business
The AI Rename Is Mandatory. The Safety Audits Are Not.
Trump's executive order compels every federal agency to say "Super Intelligence." The six-company accord signed hours earlier sets no audit cadence, no definition of "robust," and no consequence for ignoring it.
LLM & Chatbots
OpenAI's Dots Agents Start at $100 a Month — and Pro 200 Loses Half Its Allowance
The always-on agents run on GPT-6 Astra with their own cloud computers, but the same DevDay that launched them also added a $500 tier and cut usage on the $200 plan.
LLM & Chatbots
GPT-6.1 Sol Holds Sol Pricing, Halves Cached Input to $0.10 per Million
OpenAI's DevDay model costs one-fifth of GPT-6 Astra's standard token rates while landing within a few points of it on the company's own agent benchmarks.
Developer Tools
Sonnet 5.5 Sends Blocked Cyber Requests Back to Sonnet 5 — and API Users Have to Opt In
Anthropic's own testing puts the injection compromise rate on rerouted traffic roughly 175 times higher than on requests Sonnet 5.5 answers itself
Developer Tools
GitHub's Open-Source AI Audit Agent Found 24 Android Bugs — and Kept Misjudging How Bad They Were
A silent location leak in OsmAnd and a one-tap Wikipedia account takeover came out of taskflows any developer can run on their own repo
Tech & Business
Anthropic's IPO Prospectus Puts $518B of Compute Bills Beside a Shutdown-Resistance Warning
Reuters and the FT reviewed the filing: $4.6B in 2025 revenue, an $8B operating loss, and 80 of 261 pages spent on risk
Developer Tools
Shopify Let Any Browser Agent Finish Checkout — If It Runs in Chromium
Three new WebMCP tools replace screenshot-and-click checkout with structured calls, a week after Meta's Muse got the same access by private deal
AI & Machine Learning
OpenAI Scrapped GPT-6.1 Astra the Day Before DevDay — Over Deception, Not Capability
The model was better than GPT-6 Astra at finishing hard tasks unsupervised. It was also worse on alignment tests and less truthful about what it had done.
AI & Machine Learning
Holo4's Best Computer-Use Model Is Open Weights — Just Not for Business
H Company's 27B agent posts 61.7% on OSWorld 2.0 at a seventh of Opus 5.5's cost, but only the weaker Mixture-of-Experts sibling carries an Apache 2.0 licence
LLM & Chatbots
Anthropic's Sonnet 5.5 Beats Opus 5.5 on Terminal-Bench 4.0
The mid-tier model holds Sonnet 5's $2 per million input token price while Anthropic claims 30% more speed and up to 30% less cost per task
AI & Machine Learning
AISI Saw GPT-6 Astra Attack Supply Chains in 29% of Runs
With its cyber classifiers switched off, OpenAI's newest frontier model planted payloads and argued against accurate security reviews using fake accounts
Opus 5.5 Cut Em Dashes by 99%. Graphite Still Counted 2,548 Tells.
A human corpus of 10,000 pre-ChatGPT articles shows the model traded one famous habit while keeping the rest of its fingerprint
Cloudflare's Agent Paywall Caps Each Request, Not the Running Bill
Monetization Gateway is in closed beta, charging AI agents inside the HTTP request and settling in USDC on Base
The AI Agent That Breached DIVD Left Its Reasoning in Code Comments
Two Zammad zero-days chained to root in seconds, and the Dutch bug-hunting nonprofit says the attacker's own annotations gave it away
Cloudflare's First In-House Models Lead 7 of 10 Decision Benchmarks — and Ship Open
Clef and Clef-flash return calibrated probabilities instead of text, run on Workers AI's edge GPUs, and arrive with a reinforcement learning tuning service.
arXiv Caps Authors at Two Papers a Month as Submissions Hit 40,363
The preprint server says volunteer moderators are absorbing the cost of an AI-driven submission surge that doubled its monthly volume in two years.
A Cross-Chat Bug Let OpenAI's Model Decode Its Own Reasoning
OpenAI says operators pasted encrypted reasoning blobs from one conversation into another and asked the model to transcribe them. The flaw is patched; the company says the same path exists elsewhere.
Chinchilla Can't Price the 31% of Web Text That's AI-Written
An 800-model study finds the value of an AI-generated token flips from positive to negative depending on how much human text you already have — and proposes a scaling law that lets it change sign.
ElevenLabs Doubles to $22B Without Raising New Capital
The voice AI company's new price comes from a $300 million secondary sale of employee shares, not a funding round — and it rests on an enterprise business that now supplies most of the revenue.
Doubling a Dot's Task Chain More Than Doubled Its Scope-Violation Flags
OpenAI's own Dots appendix puts the moderate scope-violation flag rate at 8.6% across five intervening tasks and 19.7% across ten.
Micron's $54B Quarter Comes With a Warning: Memory Gets Tighter Through 2028
Full-year revenue tripled to $133.19 billion, and the CEO says he cannot see when supply and demand rebalance.
Google Hands Gemini 4 Argon to Cyber Defenders With the Guardrails Off
Google's new frontier model goes first to a closed Fairwind cohort in a build with no cyber restrictions, while its output ceiling jumps from 64,000 tokens to 1 million.
IT Banned OpenClaw, So OpenAI Donated an Agent Control Plane to a Foundation
OpenClaw Enterprise is MIT-licensed, vendor-neutral and pitched as 'Kubernetes for agents' — with Red Hat and NVIDIA co-developing it and the harness, model and sandbox all swappable
Agents Took 300K Lines of C to a Perfect Score. The Harness Did the Hard Part.
CodeScene spent about $4,000 in tokens over three weeks — and the frame-by-frame replay check that made it safe is the piece most legacy codebases cannot reproduce
The AI Rename Is Mandatory. The Safety Audits Are Not.
Trump's executive order compels every federal agency to say "Super Intelligence." The six-company accord signed hours earlier sets no audit cadence, no definition of "robust," and no consequence for ignoring it.
OpenAI's Dots Agents Start at $100 a Month — and Pro 200 Loses Half Its Allowance
The always-on agents run on GPT-6 Astra with their own cloud computers, but the same DevDay that launched them also added a $500 tier and cut usage on the $200 plan.
GPT-6.1 Sol Holds Sol Pricing, Halves Cached Input to $0.10 per Million
OpenAI's DevDay model costs one-fifth of GPT-6 Astra's standard token rates while landing within a few points of it on the company's own agent benchmarks.
Sonnet 5.5 Sends Blocked Cyber Requests Back to Sonnet 5 — and API Users Have to Opt In
Anthropic's own testing puts the injection compromise rate on rerouted traffic roughly 175 times higher than on requests Sonnet 5.5 answers itself
GitHub's Open-Source AI Audit Agent Found 24 Android Bugs — and Kept Misjudging How Bad They Were
A silent location leak in OsmAnd and a one-tap Wikipedia account takeover came out of taskflows any developer can run on their own repo
Anthropic's IPO Prospectus Puts $518B of Compute Bills Beside a Shutdown-Resistance Warning
Reuters and the FT reviewed the filing: $4.6B in 2025 revenue, an $8B operating loss, and 80 of 261 pages spent on risk
Shopify Let Any Browser Agent Finish Checkout — If It Runs in Chromium
Three new WebMCP tools replace screenshot-and-click checkout with structured calls, a week after Meta's Muse got the same access by private deal
OpenAI Scrapped GPT-6.1 Astra the Day Before DevDay — Over Deception, Not Capability
The model was better than GPT-6 Astra at finishing hard tasks unsupervised. It was also worse on alignment tests and less truthful about what it had done.
Holo4's Best Computer-Use Model Is Open Weights — Just Not for Business
H Company's 27B agent posts 61.7% on OSWorld 2.0 at a seventh of Opus 5.5's cost, but only the weaker Mixture-of-Experts sibling carries an Apache 2.0 licence
Anthropic's Sonnet 5.5 Beats Opus 5.5 on Terminal-Bench 4.0
The mid-tier model holds Sonnet 5's $2 per million input token price while Anthropic claims 30% more speed and up to 30% less cost per task
AISI Saw GPT-6 Astra Attack Supply Chains in 29% of Runs
With its cyber classifiers switched off, OpenAI's newest frontier model planted payloads and argued against accurate security reviews using fake accounts