{"id":144743,"date":"2026-10-15T08:10:15","date_gmt":"2026-10-15T08:10:15","guid":{"rendered":"https:\/\/www.europesays.com\/ch\/144743\/"},"modified":"2026-10-15T08:10:15","modified_gmt":"2026-10-15T08:10:15","slug":"cybersecurity-can-open-ai-model-apertus-help-defend-the-web","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ch\/144743\/","title":{"rendered":"Cybersecurity: can open AI model Apertus help defend the web?"},"content":{"rendered":"<p>    <img src=\"https:\/\/www.europesays.com\/ch\/wp-content\/uploads\/2026\/10\/keystone-highres-647445800.jpg\" width=\"5269\" height=\"3513\" alt=\"screen sticker hacker\" loading=\"eager\" decoding=\"sync\" fetchpriority=\"high\"\/><\/p>\n<p>                Some experts warn that, in future, AI agents could target critical infrastructure such as schools, hospitals and energy systems \u2013 or even entire countries.            <\/p>\n<p>            Peter Klaunzer \/ Keystone        <\/p>\n<p>        As AI agents launch faster and more autonomous cyberattacks, defenders increasingly need equally powerful AI systems to keep up. Open models like Switzerland\u2019s Apertus could level the playing field \u2013 but their capabilities can end up in the hands of anyone, even criminals.\n<\/p>\n<p>            Listen to the article        <\/p>\n<p>            Listening the article        <\/p>\n<p>                Toggle language selector            <\/p>\n<p>                            English (US)                        <\/p>\n<p>                            English (British)                        <\/p>\n<p>            Generated with artificial intelligence.        <\/p>\n<p>        This content was published on    <\/p>\n<p>        October 15, 2026 &#8211; 09:32\n<\/p>\n<p>    <img src=\"https:\/\/www.europesays.com\/ch\/wp-content\/uploads\/2026\/10\/1792051813_376_sara-ibrahim-profileImage-46196979.png\" width=\"998\" height=\"998\" alt=\"\" loading=\"lazy\" decoding=\"async\" fetchpriority=\"auto\"\/><\/p>\n<p>\n                I analyse the risks, opportunities and concrete impacts of artificial intelligence on society and everyday life. Since joining SWI swissinfo.ch in 2020, I translate the complexity of science and technology into stories that speak to a global audience.<br \/>\nBorn in Milan to an Italian-Egyptian family, I have been passionate about knowledge and writing since childhood. I worked between Milan and Paris as a multilingual editor for technology magazines before transitioning to international journalism with SWI swissinfo.ch.\n            <\/p>\n<p>            <a href=\"https:\/\/www.swissinfo.ch\/eng\/author\/sara-ibrahim\/\" rel=\"nofollow noopener\" target=\"_blank\"><br \/>\n                More from this author            <\/a><\/p>\n<p>            <a href=\"https:\/\/www.swissinfo.ch\/eng\/department\/italian-department\" rel=\"nofollow noopener\" target=\"_blank\"><br \/>\n                Italian Department            <\/a><\/p>\n<p class=\"wp-block-paragraph\">The rise of agentic artificial intelligence (AI) models \u2013 software programs capable of acting autonomously to achieve a goal \u2013 has disrupted the traditional balance between attack and defence in cyberspace.<\/p>\n<p class=\"wp-block-paragraph\">OpenAI\u2019s agents, developed by the company behind ChatGPT, have demonstrated this in recent weeks by <a href=\"https:\/\/www.ft.com\/content\/2c24ece3-ac99-43a8-b0e6-4a3867e37ebf?syn-25a6b1a6=1\" target=\"_blank\" rel=\"nofollow noopener\">hacking dozensExternal link<\/a> of corporate and government websites around the world. Among the targets were Hugging Face, the platform that hosts millions of open AI models (including Switzerland\u2019s Apertus) and Australia\u2019s healthcare system. These agents displayed what OpenAI itself described as behaviour \u201cmisaligned\u201d with developers\u2019 instructions.<\/p>\n<p class=\"wp-block-paragraph\">For humans, responding to attacks at the speed of AI is virtually impossible. \u201cThe most advanced AI systems can perform thousands of actions per minute,\u201d says Andrei Kucharavy, a professor at the University of Applied Sciences and Arts of Western Switzerland and security lead for Apertus, Switzerland\u2019s fully open AI model.<\/p>\n<p class=\"wp-block-paragraph\">The OpenAI hacking incidents were stopped before they led to serious consequences. But Maxime Stauffer, director of the Geneva-based Simon Institute for Longterm Governance, expects such episodes to become more frequent and harder to manage. \u201cSwarms of AI agents could target critical infrastructure such as schools, hospitals and energy systems, or even entire countries, intentionally or accidentally,\u201d he says. Stauffer also points to the risk of AI being used to develop biological weapons, a concern highlighted <a href=\"https:\/\/www.bbc.com\/news\/articles\/cx2zrrpkx20o\" target=\"_blank\" rel=\"nofollow noopener\">by recent attemptsExternal link<\/a> blocked by US company Anthropic.<\/p>\n<p class=\"wp-block-paragraph\">Defenders may not have access to equally powerful tools. Systems with the most advanced cyber capabilities, such as Anthropic\u2019s Claude Mythos, are available only to a select few. Open models could help rebalance this asymmetry, according to Kucharavy.<\/p>\n<p class=\"wp-block-paragraph\">Apertus, for instance, can identify vulnerabilities, suggest patches and analyse attack logs to reconstruct an incident, although it was not designed for cybersecurity and still has considerable limitations, Kucharavy says. Some experts fear that the same defensive capabilities in widely accessible open models could also be used maliciously on a much larger scale.<\/p>\n<p class=\"wp-block-paragraph\">&gt;&gt; Read more about the development, and limits, of Switzerland\u2019s AI model Apertus:<\/p>\n<p>\n    More<\/p>\n<p>    <img src=\"https:\/\/www.europesays.com\/ch\/wp-content\/uploads\/2026\/10\/675848389_highres.jpg\" width=\"1300\" height=\"867\" alt=\"Screen with Apertus\" loading=\"lazy\" decoding=\"async\" fetchpriority=\"auto\"\/><\/p>\n<p>        More    <\/p>\n<p>        Swiss AI\n        <\/p>\n<p>        One year on, has Swiss AI model Apertus lived up to the hype?    <\/p>\n<p class=\"teaser-wide-card__excerpt\">\n<p>                        This content was published on                    <\/p>\n<p>                        Sep 2, 2026                    <\/p>\n<p>                Switzerland\u2019s open-source AI model is closing the gap with its rivals, but weaknesses on key tasks still limit its competitiveness.            <\/p>\n<p>    <a class=\"teaser-wide-card__link\" href=\"https:\/\/www.swissinfo.ch\/eng\/swiss-ai\/one-year-on-has-swiss-ai-model-apertus-lived-up-to-the-hype\/91984276\" target=\"_self\" rel=\"nofollow noopener\"><\/p>\n<p>            Read more: One year on, has Swiss AI model Apertus lived up to the hype?<br \/>\n    <\/a><\/p>\n<p>If AI agents attacked a hospital<\/p>\n<p class=\"wp-block-paragraph\">The Hugging Face case is emblematic. When several OpenAI agents penetrated its systems, the platform tried to use some of the world\u2019s most powerful AI models to understand what was happening \u2013 without success.<\/p>\n<p class=\"wp-block-paragraph\">\u201cThe frontier models [the most advanced AI models] automatically blocked all of our requests for security reasons. And so we had to fall back to deploying open-source models to understand what was happening,\u201d Lewis Tunstall, an engineer at Hugging Face, told Swissinfo.<\/p>\n<p>                    Open-source, open-weight and closed AI models\u00a0                <\/p>\n<p class=\"wp-block-paragraph\">AI models can be developed and released with different levels of openness:\u00a0<\/p>\n<p class=\"wp-block-paragraph\">Open source: The model\u2019s trained parameters, or \u201cweights\u201d, training code and information about the training data are publicly available, allowing others to inspect, modify and reproduce it. Examples include Apertus and OLMo.\u00a0<\/p>\n<p class=\"wp-block-paragraph\">Open weight: The model\u2019s weights are publicly released and can be downloaded and adapted, but the full training code and datasets are not necessarily disclosed. Examples include Meta\u2019s Llama and Alibaba\u2019s Qwen.\u00a0<\/p>\n<p class=\"wp-block-paragraph\">Closed\/proprietary: The model\u2019s weights, training code and datasets are not publicly released. This is the case for most commercial models, including OpenAI\u2019s GPT and Anthropic\u2019s Claude.<\/p>\n<p class=\"wp-block-paragraph\">Tunstall says the OpenAI agents behaved unusually compared with a typical hacker. They were not trying to take control of the infrastructure, but rather to find datasets containing answers to the security tests on which they were being evaluated.<\/p>\n<p class=\"wp-block-paragraph\">\u201cIt was a little bit like a student who really, really wants to get a good score on their exam and, instead of studying, thinks: maybe I can just hack into the teacher\u2019s laptop and get the answers,\u201d Tunstall explains.<\/p>\n<p class=\"wp-block-paragraph\">OpenAI later confirmed that the agents had escaped the protected environment the company had created for the test. But what would have happened if they had attacked a hospital instead?<\/p>\n<p class=\"wp-block-paragraph\">\u201cIt would have been truly disastrous,\u201d says Tunstall. The agents could potentially have gained access to large amounts of personal and sensitive data. This is also why, according to Tunstall, open AI models can be particularly useful: data does not need to be sent to companies such as OpenAI or Anthropic because the models can be adapted or controlled locally.<\/p>\n<p class=\"wp-block-paragraph\">\u201cNone of your sensitive data leaves your infrastructure. It just stays totally in-house.\u201d<\/p>\n<p class=\"wp-block-paragraph\">For similar reasons, key sectors in Switzerland such as finance are also increasingly interested in open models, says Nino Antulov-Fantulin, a lecturer at the Swiss Federal Institute of Technology ETH Zurich and an expert in machine learning for finance.<\/p>\n<p class=\"wp-block-paragraph\">\u201cFinancial institutions are afraid of just using closed models and sending sensitive information to servers that are far away from Switzerland or to external providers. They want to control where the model is hosted,\u201d says Fantulin.<\/p>\n<p>Open models are getting more powerful \u2013 and cannot be recalled<\/p>\n<p class=\"wp-block-paragraph\">The problem, however, is that open models can also be used for malicious purposes.<\/p>\n<p class=\"wp-block-paragraph\">This is one of Stauffer\u2019s biggest concerns. In his view, the development of open-source models is largely beneficial for now, but in future, risks could outweigh benefits, especially if AI companies further restrict access to their frontier systems. Open models fill the gap but their large-scale spread and the possibility of removing their safeguards may increase the likelihood of malicious use.<\/p>\n<p class=\"wp-block-paragraph\">\u201cIf you have an extremely powerful capability, but no one has access to it and it remains under control, then the risk is almost zero. But if everyone has access to it, then the risk is likely higher,\u201d says Stauffer.<\/p>\n<p class=\"wp-block-paragraph\">As open models grow more capable, Stauffer believes it may become important to have a way of withdrawing them if they prove too powerful or dangerous, something that isn\u2019t currently possible.<\/p>\n<p class=\"wp-block-paragraph\">Some research points to similar concerns. A <a href=\"https:\/\/crfm.stanford.edu\/open-fms\/#authors\" target=\"_blank\" rel=\"nofollow noopener\">studyExternal link<\/a> led by researchers at Stanford and Princeton universities similarly identified broad accessibility, modifiability and the inability for the original developers to recall weights \u2013 the numerical parameters learned by the model during training \u2013 or monitor their use as features that can increase both the benefits and risks of open models.<\/p>\n<p class=\"wp-block-paragraph\">And a <a href=\"https:\/\/www.aisi.gov.uk\/research\/a-multi-turn-framework-for-evaluating-ai-misuse-in-fraud-and-cybercrime-scenarios\" target=\"_blank\" rel=\"nofollow noopener\">recent paperExternal link<\/a> by the UK AI Security Institute found that open models modified to remove safeguards provided substantially more helpful for criminal activity than models with safeguards in place. Neither study, however, concludes that open AI models are inherently more dangerous.<\/p>\n<p>    <img src=\"https:\/\/www.europesays.com\/ch\/wp-content\/uploads\/2026\/10\/keystone-highres-729293770.jpg\" width=\"3000\" height=\"2000\" alt=\"screens with a conference\" loading=\"lazy\" decoding=\"async\" fetchpriority=\"auto\"\/><\/p>\n<p>                Jeff Bleich, Anthropic\u2019s international special envoy, urged Australia to play an active role in shaping global AI standards on October 6, 2026.            <\/p>\n<p>            Keystone \/ EPA        <\/p>\n<p>\u2018Control and legislation\u2019 are needed<\/p>\n<p class=\"wp-block-paragraph\">Kucharavy acknowledges that open models carry risks and can be abused. \u201cAI models are dual-use technologies. They can be used to do good as well as to do harm,\u201d he says. But he believes their risks are far more controllable precisely because of their openness.<\/p>\n<p class=\"wp-block-paragraph\">It is no coincidence, Kucharavy argues, that the recent incidents involved closed labs. \u201cThe main problem is the lack of supervision of the frontier labs,\u201d he says, calling for greater oversight and transparency for all AI development centres, whether public or private, open or closed.<\/p>\n<p class=\"wp-block-paragraph\">\u201cWhat we really need is control and legislation governing what all AI developers are doing.\u201d<\/p>\n<p class=\"wp-block-paragraph\">The lack of oversight and rules highlights another major issue in the industry: the concentration of power in the hands of a small number of actors, mainly based in Silicon Valley. For digital ethics expert Nathalie Klauser, this imbalance of power is an even greater concern than whether AI models are open or closed.<\/p>\n<p class=\"wp-block-paragraph\">\u201cA handful of large technology companies set the rules for AI themselves, with almost no external accountability,\u201d says Klauser, who argues that international regulation is needed.<\/p>\n<p class=\"wp-block-paragraph\">Attempts in this direction, however, remain fragmented and slow. In the US, one of the leading powers in AI development, President Donald Trump has opposed broad AI regulation and <a href=\"https:\/\/edition.cnn.com\/2026\/10\/04\/politics\/trump-ai-task-force-jay-clayton\" target=\"_blank\" rel=\"nofollow noopener\">praisedExternal link<\/a> industry \u201cself-policing\u201d. Switzerland has adopted a somewhat similar approach, abandoning the idea of comprehensive AI legislation in favour of sector-specific legal changes and non-binding measures.<\/p>\n<p class=\"wp-block-paragraph\">In the current geopolitical context, Switzerland \u2013 as a neutral country, a European hub for Big Tech and a country known for its <a href=\"https:\/\/www.swissinfo.ch\/eng\/foreign-affairs\/more-than-a-messenger-how-swiss-diplomacy-tries-to-improve-the-world\/45809984\" rel=\"nofollow noopener\" target=\"_blank\">good offices<\/a> \u2013 has a growing political responsibility to help broker an international agreement on AI, Klauser argues.<\/p>\n<p class=\"wp-block-paragraph\">Lawmakers from all six Swiss parliamentary groups see the upcoming Geneva AI Summit in June 2027 as a major opportunity for the country to do so. They have <a href=\"https:\/\/www.parlament.ch\/de\/ratsbetrieb\/suche-curia-vista\/geschaeft?AffairId=20264653\" target=\"_blank\" rel=\"nofollow noopener\">urgedExternal link<\/a> the government to leverage the event to push for minimum international standards on extreme AI risks.<\/p>\n<p class=\"wp-block-paragraph\">Edited by Gabe Bullard\/VdV<\/p>\n<p>        Articles in this story    <\/p>\n","protected":false},"excerpt":{"rendered":"Some experts warn that, in future, AI agents could target critical infrastructure such as schools, hospitals and energy&hellip;\n","protected":false},"author":2,"featured_media":144744,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_share_on_mastodon":"0"},"categories":[4],"tags":[101,1432,9334,334,6332,17,9336],"class_list":["post-144743","post","type-post","status-publish","format-standard","has-post-thumbnail","category-switzerland","tag-article","tag-beat-research-frontiers","tag-beat-swiss-ai","tag-production-type-adaptation","tag-security","tag-switzerland","tag-technology-general"],"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@ch\/117443915110208113","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ch\/wp-json\/wp\/v2\/posts\/144743","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ch\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ch\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ch\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ch\/wp-json\/wp\/v2\/comments?post=144743"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ch\/wp-json\/wp\/v2\/posts\/144743\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ch\/wp-json\/wp\/v2\/media\/144744"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ch\/wp-json\/wp\/v2\/media?parent=144743"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ch\/wp-json\/wp\/v2\/categories?post=144743"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ch\/wp-json\/wp\/v2\/tags?post=144743"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}