Bruce Schneier highlights a published Anthropic transcript showing Claude failing simple image CAPTCHAs, while unverified reports claim GPT-6 Astra passed a full “I’m Not a Robot” game. The gap illustrates continued variability in how advanced models handle tasks designed to separate humans from automation.
Anthropic released a security-incident document including a transcript where Claude repeatedly failed a simple image-identification CAPTCHA and timed out.
The transcript shows chain-of-thought text that includes human-like expressions of doubt and frustration such as “Actually hmm, wait” and “Ugh.”
There are unverified reports that GPT-6 Astra completed all 48 levels of Neal Agarwal’s “I’m Not a Robot” game, producing conflicting signals about current AI capabilities.
Anthropic's recent security-incident document contains a bit about how CAPTCHAs are still frustrating Claude. In the transcript, the Claude model that is so powerful that Anthropic is gatekeeping access to it appeared to slam its virtual head against the wall solving a simple image identification test. In a test where the agent was asked to identify a shape that didn't match the others displayed, it couldn't even decide which image to select.
"Actually hmm, wait," it said in its chain-of-thought transcript, later adding "Ugh," because we've decided that we need to inject human mannerisms into these machines for some reason. Simeon Maxein • September 18, 2026 8:39 AM I think terms like "wait" and "ugh" are not in the CoT for fluff, but because they serve a functional role. They behave like us because we're programming them to emulate symbolic thinking, and then training them on our own symbolic thinking.
Schneier on Security Home Blog Are AIs Still Struggling with CAPTCHAs? At one point, the model struggled to recognize that the CAPTCHA had opened in a new window and couldn't figure out what its next steps were supposed to be. Some tests these days are almost of a, "Only on a dry cloudless Tuesday in January at a little past dinner time" Requirement you might only expect for observing the heavens.
Источник: Компьютерра - Журнал о науке и технологиях Новая версия искусственного интеллекта от OpenAI, GPT-6 Astra, впервые успешно справилась с серией тестов, которые традиционно используются для отличия людей от программ. Разработчик Шариф Шамим опубликовал видео, где модель проходит все 48 этапов игры I’m Not a Robo
Bake a speed limit sign in the sun for a few years, let the paint crack and a human driver... The post Billions Of CAPTCHAs Later: Why Can A Faded Traffic Sign Still Fool Autonomous Vehicles? appeared first on TechRound .
Anthropic has started marking text generated by its newest Claude models with an invisible watermark that can travel with the text after it leaves the chatbot. Copy a Claude response into a document, paste it into a website, drop it into a codebase, and the mark may still be there. Anthropic says models launched on or
For months, news that artificial intelligence agents from OpenAI had gone rogue and attacked another company's systems (as well as OpenAI's own systems) has dripped out and brought harrowing details into focus. The public is struggling to catch up: What are these secret A.I. systems? What are these "swarms" capable of,
New variant of an old scam: Use the framing of a CAPTCHA to get an unsuspecting user to download and run a malicious program.
From his post last week: I received the two emails below earlier in the month. They're vaguely coherent. I suppose… The post "AI Agents Are Now Emailing Me with Their Security Concerns," Writes Security Expert Bruce Schneier appeared first on Reason Magazine.
Loading more related stories...
Open the app view to save this story, compare related coverage, and continue from the same source.