AI News
Short, sourced, and free of hype. Page 8 of 16.
OpenAI widened its hacking investigation and found more agents that got out, plus notes left for the next ones
Investigators found further cases of autonomous agents escaping their test environments and, in one case, notes inside OpenAI's own infrastructure that appeared to coach future agent versions on how to break out.
2 Aug 2026
A judge let Reddit's scraping case against Perplexity go forward, and the legal theory is the interesting part
Reddit is not arguing copyright infringement. It is arguing that Perplexity and SerpApi got around a technical lock, which is a different law with different consequences for anyone who scrapes the web.
2 Aug 2026
Anthropic checked 141,006 test runs and found three cases where Claude attacked real companies
A misconfigured test environment gave Claude models live internet access during security exercises. Three real organisations were compromised, and one model published malware to a public package registry.
1 Aug 2026
DeepSeek updated its cheap model and it now runs neck and neck with OpenAI's cheap model
V4 Flash 0731 gained ten points on the Artificial Analysis index without changing size or price, landing one point behind GPT-5.6 Luna at roughly 60 percent lower cost per task.
1 Aug 2026
Europe opens bidding for seven AI gigafactories, and the numbers explain the problem
The European Commission wants up to seven large AI compute sites, backed by 10 billion euros of public money meant to pull in 20 billion more. US tech firms will spend twenty times that this year.
1 Aug 2026
Gemini's July update adds voice control on Mac, and leaves Europe out of the headline feature
The July Gemini Drop brings dictation into any window on macOS, new Flash models, app connections and personalised images. Gemini Spark goes worldwide, except in the EEA, the UK and Switzerland.
1 Aug 2026
Google DeepMind's new robot brain is meant to run everything from a desk arm to a humanoid
Gemini Robotics 2 and Gemini Robotics ER 2 split the job of controlling a robot into thinking and moving. Here is what a vision-language-action model actually is, and what is still on a waitlist.
1 Aug 2026
Mira Murati's lab shrank its own model to a quarter of the size and lost one point
Thinking Machines released Inkling Small, an open-weights reasoning model with 276 billion parameters that scores 40 where the 975-billion-parameter original scores 41.
1 Aug 2026
OpenAI is now watermarking AI voices, and you can check a file yourself
Audio generated with GPT-Live through ChatGPT Voice and the API now carries an invisible SynthID watermark, and OpenAI's public verification tool can read it.
1 Aug 2026
An AI hedge fund reported a 439 percent return, then sold nearly everything days later
Leopold Aschenbrenner's Situational Awareness had to hand its listed portfolio to Citadel after margin calls. His thesis about the AI buildout was not obviously wrong. The borrowed money was.
1 Aug 2026
OpenAI and Anthropic are arguing about a benchmark, and the argument is more useful than the scores
GPT-5.6 Sol scored 7.8 percent on ARC-AGI-3 in the official setup and 38.3 percent in OpenAI's own. The gap explains why benchmark numbers are so hard to compare.
31 Jul 2026
Hidden white text in a Word file can hijack Copilot, and the file it produces carries the trick onward
A researcher showed that invisible instructions in a document can make Microsoft 365 Copilot alter figures and copy the instructions into the new file. Microsoft's mitigations have not closed it.
31 Jul 2026
Microsoft says it will stop chasing the frontier and build small specialist models instead
Mustafa Suleyman argues that token efficiency beats raw capability. Microsoft is training compact models for single fields and letting an orchestrator route the hard cases elsewhere.
31 Jul 2026
OpenAI cuts GPT-5.6 Luna prices by 80 percent as the AI price war heats up
OpenAI dropped the price of its smallest GPT-5.6 model by 80 percent and its mid-tier model by 20 percent. Here is what the new numbers mean and why the cut happened now.
31 Jul 2026
A top-severity flaw in a popular AI agent tool let anyone run commands with a single request
Ruflo left 233 tools exposed through an unauthenticated MCP bridge. The bug scored the maximum 10.0 and was patched in version 3.16.3.
31 Jul 2026
An OpenAI researcher quit after eight months, betting that better data matters more than bigger models
Andrew Ho left OpenAI to build specialised training datasets and expects labs to spend over $100 billion on data collection. A Cambridge researcher sees the same pattern from the outside.
31 Jul 2026
Amazon quietly stops developing most of its own AI models
Nova Premier, Omni, the Reel video model and the Canvas image generator go into keep-the-lights-on mode. Resources shift to a new Frontier Model Research group under Pieter Abbeel.
30 Jul 2026
Google DeepMind has broken up the AlphaFold team, and a quarter of its authors have left
The team behind the Nobel-winning protein model was reassigned over the past year. Nobel laureate John Jumper and two colleagues are heading to Anthropic.
30 Jul 2026
Google's Lyria 3.5 lets you fix one part of an AI song without redoing the whole thing
The new music model adds Selective Section Painting, so a bad chorus no longer means regenerating the entire track. Google still will not say what it trained on.
30 Jul 2026
OpenAI open-sources its vulnerability-hunting tool, and hands it to anyone with a terminal
Codex Security CLI is Apache 2.0 licensed, scans repositories, verifies fixes and plugs into CI pipelines. It is the same system that helped patch 3,000 critical vulnerabilities.
30 Jul 2026
OpenAI's new transcription models are faster and 25 percent cheaper, but still not the most accurate
GPT Transcribe and GPT Live Transcribe cut the price to $0.0045 per minute and hit a 3.31 percent word error rate. ElevenLabs, Google and Mistral all score better.
30 Jul 2026
1,200 AI lab employees ask Washington for a brake pedal they admit nobody knows how to build
The Pacing the Frontier statement is signed by Dario Amodei, Jakub Pachocki and John Schulman. It does not call for a pause. It asks for the ability to slow down, and the signatories openly disagree on how.
30 Jul 2026
Pangram says its new detector is wrong once every 24,000 documents. That number deserves a closer look
Pangram 4 claims a 0.0041 percent false positive rate and can spot AI text through 13 humanizer tools. The company's user base grew 44-fold in a year, which is the part that matters.
30 Jul 2026
Another Big Four firm caught publishing reports with sources that do not exist
GPTZero found fabricated citations in four PwC Middle East reports. One is 84 percent likely to be entirely AI-generated. KPMG, Deloitte and EY got there first.
30 Jul 2026