{"id":3582,"date":"2026-05-26T21:22:11","date_gmt":"2026-05-26T21:22:11","guid":{"rendered":"https:\/\/owspakistan.com\/?p=3582"},"modified":"2026-05-26T21:22:11","modified_gmt":"2026-05-26T21:22:11","slug":"d41586-026-01617-2","status":"publish","type":"post","link":"https:\/\/owspakistan.com\/?p=3582","title":{"rendered":"is Mythos the start of the restricted-AI era?"},"content":{"rendered":"
\n

In April, the artificial-intelligence firm Anthropic announced it had made an AI model too dangerous to be released to the public. The company, based in San Francisco, California, said its Claude Mythos model was so powerful that it had found vulnerabilities in every major operating system and web browser currently in use. \u201cThe fallout \u2014 for economies, public safety, and national security \u2014 could be severe,\u201d the company stated in a blogpost about Project Glasswing, its name for the limited release of the model to a group of 50 or so trusted organizations.<\/p>\n

The decision marks a turn to secretive, cutting-edge AI research that could become a trend, experts say. What Anthropic has done to throttle Mythos\u2019s release is likely to be adopted by other AI laboratories, reckons Helen Toner, interim executive director at the Center for Security and Emerging Technology at Georgetown University in Washington DC. \u201cI would expect this to more be the first in a series rather than a one-off,\u201d says Toner, who previously sat on the board of Anthropic\u2019s competitor OpenAI, which is also based in San Francisco.<\/p>\n

\"\"<\/p>\n

AI can design viruses, toxins and other bioweapons. How worried should we be?<\/p>\n

<\/a><\/article>\n<\/p>\n

\u201cI expect other providers to adopt a similar strategy,\u201d agrees Vasilios Mavroudis, an AI-safety researcher at the Alan Turing Institute in London. Indeed, OpenAI followed up just a week after Mythos was announced with a limited release of a cybersecurity-specific model, GPT-5.4-Cyber, to vetted researchers and organizations alone.<\/p>\n

If this kind of restricted-access AI does take off, it will mark a turning point in the long-standing debate on the merits of \u2018closed\u2019 and \u2018open\u2019 AI software \u2014 with potential knock-on implications for science. For years, researchers have argued that transparency around AI models<\/a> benefits both AI research and science in general, because researchers can study and build on the algorithms.<\/p>\n

Now there\u2019s a prospect that the makers of cutting-edge AI models might not release them widely at all. And if governments decide that powerful AI is a \u2018dual-use\u2019 technology \u2014 that is, one that could be weaponized by the military as well as used in civilian society \u2014 then extra controls, of the kind used for defence-relevant technologies, might also kick in. This could limit who gets to use the most powerful software, says Toner.<\/p>\n

Why restrict access?<\/b><\/h2>\n

Firms have tried restricting models before. In February 2019, OpenAI released a cut-down version of its GPT-2 model, citing fears it could be misused, before allowing full access that November. But, viewed by today\u2019s standards, that model had very little capability \u2014 it could complete rudimentary sentences.<\/p>\n

It\u2019s difficult for researchers who lack access to Mythos to know whether all of Anthropic\u2019s fears are well-founded. But Ciaran Martin, a management researcher at the University of Oxford, UK, who is the former chief executive of the UK National Cyber Security Centre in London, says that Mythos seems to be a \u201cbig deal\u201d and \u201ca rapid acceleration of AI capabilities\u201d.<\/p>\n

\"\"<\/p>\n

How to vibe code in science: early adopters share their tips<\/p>\n

<\/a><\/article>\n<\/p>\n

AI labs already implement \u2018guard rails\u2019 in their models to prevent misuse, usually in the form of refusals to engage with or answer queries that seem dangerous. However, those are often as simple as a set of hidden instructions given to an AI system on how to respond. They can be unwound (or \u2018jailbroken\u2019) if users so choose.<\/p>\n

The risk in giving everyone equal access to Mythos is that it could help attackers first, says Mavroudis. \u201cDefenders can use these to find security issues in systems or software projects. Attackers can do the same,\u201d he says. That is why Anthropic is giving defenders a head start. Some unauthorized access to Mythos has reportedly occurred already, however.<\/p>\n

The firm has said that its \u201ceventual goal\u201d is to enable users \u201cto safely deploy Mythos-class models at scale\u201d, but didn\u2019t respond to Nature<\/i>\u2019s query about whether this meant the public would have access. OpenAI has also been unclear, saying that it was \u201cstarting\u201d with a limited release of GPT-5.4-Cyber, and a quickly released follow-up model, GPT-5.5-Cyber; it has since launched a cybersecurity-focused product, called Daybreak, which is built on these tools. However, for now, the \u2018Cyber\u2019 models are available only to authorized users.<\/p>\n

When asked whether its models might become publicly available, an OpenAI spokesperson pointed to language in a blogpost announcing GPT-5.5-Cyber. It states that \u201cexpanding access [\u2026] responsibly requires stronger confidence in who is using the model, what systems they are targeting, and whether the work is authorized\u201d, but \u201cwe expect access to broaden over time\u201d.<\/p>\n

Where cyber goes, science follows? <\/b><\/h2>\n

Firms that restrict access to AI models aren\u2019t only concerned about cybersecurity risks. For example, companies launching models specialized for biology research have said that they\u2019re worried about potential misuse related to creating bioweapons<\/a>.<\/p>\n

In April, OpenAI released GPT-Rosalind, aimed at life scientists, which it said it would launch through a \u2018trusted-access\u2019 structure so that only approved users could explore it. The company will also monitor how the model is being used, YunYun Wang, OpenAI\u2019s life-sciences product leader, told Nature<\/i>. And last year, Google launched an \u2018AI co-scientist\u2019 system<\/a> which, again, is available only to researchers who apply for access.<\/p>\n

If such restrictions persist, they could mean that only well-connected researchers have access to the most-powerful AI tools. Researchers already worry that price hikes to existing public AI systems are starting to entrench inequality in the field, with some groups not able to afford premium subscriptions<\/a>.<\/p>\n<\/div>\n

<\/p>\n","protected":false},"excerpt":{"rendered":"

In April, the artificial-intelligence firm Anthropic announced it had made an AI model too dangerous to be released to the public. The company, based in San Francisco, California, said its Claude Mythos model was so powerful that it had found vulnerabilities in every major operating system and web browser currently in use. \u201cThe fallout \u2014<\/p>\n","protected":false},"author":1,"featured_media":3583,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2}},"categories":[5],"tags":[2231,98,711,99,97,235],"class_list":["post-3582","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-science","tag-computer-science","tag-humanities-and-social-sciences","tag-machine-learning","tag-multidisciplinary","tag-science","tag-technology"],"jetpack_publicize_connections":[],"_links":{"self":[{"href":"https:\/\/owspakistan.com\/index.php?rest_route=\/wp\/v2\/posts\/3582","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/owspakistan.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/owspakistan.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/owspakistan.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/owspakistan.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=3582"}],"version-history":[{"count":0,"href":"https:\/\/owspakistan.com\/index.php?rest_route=\/wp\/v2\/posts\/3582\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/owspakistan.com\/index.php?rest_route=\/wp\/v2\/media\/3583"}],"wp:attachment":[{"href":"https:\/\/owspakistan.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3582"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/owspakistan.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3582"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/owspakistan.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3582"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}