prompt πŸ€– AI News
13.5K subscribers
57 photos
24 videos
1 file
939 links
Welcome to @prompt, your go-to source for AI insights, breakthroughs, and tools shaping the future of intelligence.


Contact: @LightEarendil
Download Telegram
🚨 Fireworks AI gets hacked again, second breach in weeks

Bloomberg reports the Nvidia-backed inference provider has notified customers of a new incident.

β€’ The October breach exposed environment variables in eval jobs
β€’ Fireworks serves other companies' models, which means it holds their keys
β€’ Rotating credentials is the obvious move for anyone on the platform
🚨 Claude Fable 5.2 takes #1 on Artificial Analysis with a 65, and the bill doubled

Artificial Analysis ranks it the highest-scoring model so far, but that's at Max effort.

β€’ Output tokens per task jump from 118k to 274k vs Fable 5.1
β€’ Reasoning tokens alone go from 71k to 192k
β€’ At High effort it uses about 80k tokens
β€’ It beats Opus 5.5 Max at lower cost
🚨 Anthropic wants $69.3B from its IPO, which would be one of the biggest tech listings ever

Reuters reports the raise target, and it lands as a lab stops being a private bet.

β€’ Public filings mean real numbers on compute spend and margins
β€’ Quarterly earnings pressure meets a safety-first brand
β€’ OpenAI's last round valued it at $1.25T, so that's the bar
🚨 400+ "normal" AI apps hide explicit deepfake tools behind a server switch

Wired traced a connected network of apps and domains that look harmless in review, then unlock the abusive stuff later.

β€’ Disguises include pet-photo generators, language classes and AI workout coaches
β€’ A server check flips on the hidden features
β€’ One trigger was installing the app after tapping an ad on Meta platforms
🚨 CoreWeave's 2032 bond now yields over 16%

CNBC reports the 9.625% coupon paper is trading well under par, and five-year CDS hit 1,289 bps Thursday, above the October peak. Protection is also getting pricier on Nvidia, Broadcom and Oracle. Google's bonds yield around 5.7%.

Jefferies' Chris Wood says the AI capex race ends when the credit card gets taken away.
πŸ” DeepMind's AI agents just formally verified Signal's actual protocol code

DeepMind says AlphaProof Code, with Gemini Carbon agents doing the proofs, wrote over six million lines of Lean. Humans reviewed every spec. It's the first formal verification of a widely deployed protocol's implementation, with or without AI.

The agents proved the code matches the spec, but people still decided what "secure" means.
🚨 Anthropic's IPO is pricing at $1.1T to $1.3T, well short of the $2T some backers wanted

BBC reports the range is $70 to $78 a share, raising up to $64bn on Nasdaq. Revenue hit $16.2bn for the first half of the year, up from $1.1bn a year earlier.

It still lands below SpaceX, which listed in June at nearly $1.8T.
🚨 Gemini 4 Carbon is live, and Google says it's the strongest Gemini yet

Google says it ranks #1 on CursorBench 4.0, WANDR and AutomationBench at more than 2X lower cost per task than the next best external model. It burns more tokens than Argon to reason deeper.

The API is open, but in the Gemini app it looks Ultra-only. Pro subscribers are still staring at 3.1 Pro.
πŸ–ΌοΈ MiniMax opens the weights on H3-Image, a 33B model that renders 16 megapixels without an upscaler

MiniMax released one checkpoint for text-to-image and editing, distilled to 6 sampling steps with no CFG, and it takes up to 16 reference images. The DiT is 68.65 GB in BF16, and it reuses the H3 text encoder, which is a separate 66.7 GB download.
🚨 Anthropic is going after Microsoft Office, and it's doing it right before the IPO

Bloomberg reports the launch of Claude Whiteboards, Claude Sheets and Claude Data, which move brainstorming, spreadsheets and shared data into Claude itself, so enterprise files live there instead of in a chat window next to Copilot.
❀1
🚨 Stripe buys OpenRouter, and the payments company now sits in the LLM routing layer

Stripe announced that OpenRouter is now part of Stripe, framing it as economic infrastructure for AI. The stated plan is helping businesses explore new models and optimize token routing and usage, and no price or terms came with it.
πŸ€– Claude now has its own whiteboards, spreadsheets and dashboards

Anthropic launched Whiteboards and Sheets in beta on all paid plans, and Dashboards is now generally available as Claude Data. Claude builds the board or sheet with your team, then exports to Lucid, Miro or Vani by Zoho.

Miro and Lucid are listed as partners and are also the products being copied.
❀1
πŸ€– Qwen4-Max is in preview, and Alibaba still hasn't shown a single benchmark

Alibaba's Qwen team announced the new flagship on its Token Plan, with early access in QwenWork, Qoder, Qwen APP and QwenStudio. The pitch is stronger coding and agentic work, plus "deeper knowledge" and richer world understanding. No numbers, no pricing, no weights.
πŸ‘2
🚨 Australia wants to make it a crime for AI companies to fail to prevent harm from their models

The government says it will consider a new criminal offence plus a watchdog that can issue big fines, after September's OpenAI hack of the Medicare statistics portal. Nobody's defined "reasonable steps" yet, and the AFP says it has received no formal report of a crime.
🚨 Claude Code's "don't ask again" button is leaking passwords and API keys onto GitHub

Apiiro researchers found thousands of secrets, including database passwords and root logins, sitting in committed .claude/settings.local.json files, Ars Technica reports. Approved shell commands get saved as-is, secrets included, and the file is only gitignored globally, so a fresh clone or a folder that later becomes a repo can push it public.
🚨 OpenAI's revenue hits $10B, and the losses keep growing anyway

The WSJ reports that the top line is up sharply, but the bill for compute is rising faster than the checks coming in.

That's the model the whole race is priced on. Anthropic and OpenAI are both eyeing IPOs while still spending more than they bring in.
πŸ€– Gemini Robotics 2.5 put a humanoid on Mercedes-Benz production lines, and it hit 96%

DeepMind says Apptronik's Apollo 2 sorted metal brackets from bulk bins into trays in Hungary and Germany. Across 200+ tasks, success is up 30% to 100% over baselines, and tasks run 20% to 30% faster.

Safe completion around people improved 43%. It's rolling out to partners only, and a 96% pick rate on a line means a fumble every 25 parts.
🚨 A PLA-linked Chinese university bought Nvidia H200s while sitting on a US blacklist

Reuters reports that military-linked institutions found ways around the restrictions, apparently through indirect purchases rather than a front door order. A blacklist only works if the sale stops at the first buyer, and the H200 is the chip Washington said it was gatekeeping.
❀1
🚨 Grok Build leaks 35,000 vibe-coded apps to anyone who enumerates the URLs

Claroc found that the share links are guessable at scale. That exposed medical exams, Navy contractor rosters, and Tesla technician bonus data.

24 of the apps were SpaceXSI's own, including a 594-account sales territory map. The links died only after Ars emailed questions.
🧬 AlphaGenome 2 stops averaging tissues and reads DNA cell type by cell type

DeepMind trained it on 107M single-cell profiles and rebuilt it as a conditional model that predicts one cell type or state at a time. On the TenK10K immune cell map it finds causal variants 30% of the time versus 16% for the original, and directional gene effects 43% versus 23%. It also adds DNA methylation, and it's open to non-commercial research through the API today.
🚨 WSJ: Nvidia is preparing to oust Jensen Huang

The board is reportedly lining up a replacement for the man who built the AI chip empire. It's a report, not a confirmed decision.

β€’ NVDA is down roughly 8% after hours
β€’ It follows the China spy probe and a NYT story on a second affair
β€’ No successor has been named