{"id":48972,"date":"2026-07-01T12:57:56","date_gmt":"2026-07-01T19:57:56","guid":{"rendered":"https:\/\/occupysf.net\/?p=48972"},"modified":"2026-07-01T12:57:57","modified_gmt":"2026-07-01T19:57:57","slug":"anthropic-ban-not-the-ai-curb-thats-needed-security-experts-say","status":"publish","type":"post","link":"https:\/\/occupysf.net\/index.php\/2026\/07\/01\/anthropic-ban-not-the-ai-curb-thats-needed-security-experts-say\/","title":{"rendered":"Anthropic ban not the AI curb that&#8217;s needed, security experts say"},"content":{"rendered":"\n<ul>\n<li><a href=\"https:\/\/www.sfexaminer.com\/users\/profile\/Troy_Wolverton\">By Troy Wolverton | Examiner staff writer<\/a><\/li>\n\n\n\n<li>Jun 30, 2026 (SFExaminer.com)<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/bloximages.chicago2.vip.townnews.com\/sfexaminer.com\/content\/tncms\/assets\/v3\/editorial\/5\/72\/57275137-b18e-4a78-9c2b-f71170aee91a\/68e4336decd40.image.jpg?resize=400%2C267\" alt=\"Dario Amodei\"\/><figcaption class=\"wp-element-caption\"><em>Anthropic \u2014 led by CEO and co-founder Dario Amodei \u2014 closed a $13 billion venture-funding deal in September, the largest in the quarter and one of the largest ever.Jeff Chiu\/Associated Press<\/em><\/figcaption><\/figure>\n\n\n\n<p>When it comes to artificial intelligence, President Donald Trump\u2019s second administration has taken a decidedly antiregulatory stance \u2014 until this month.&nbsp;<\/p>\n\n\n\n<p>One of Trump\u2019s first moves in his second term was to&nbsp;<a href=\"https:\/\/www.sfexaminer.com\/news\/technology\/ai-companies-quiet-on-trump-rescinding-biden-safety-order\/article_26f25f72-d85f-11ef-816c-6ffae5849ee4.html\" target=\"_blank\" rel=\"noreferrer noopener\">ditch the modest AI safety guidelines issued by his predecessor<\/a>, Joe Biden. After Congress over the summer overwhelmingly&nbsp;<a href=\"https:\/\/apnews.com\/article\/congress-ai-provision-moratorium-states-20beeeb6967057be5fe64678f72f6ab0\" target=\"_blank\" rel=\"noreferrer noopener\">rejected a proposal to ban new state laws<\/a>&nbsp;<a href=\"https:\/\/apnews.com\/article\/congress-ai-provision-moratorium-states-20beeeb6967057be5fe64678f72f6ab0\" target=\"_blank\" rel=\"noreferrer noopener\">and roll back existing ones<\/a>&nbsp;governing the technology, the president issued an executive order&nbsp;<a href=\"https:\/\/www.sfexaminer.com\/news\/technology\/trump-ai-executive-order-what-it-means\/article_0751df65-06d6-4a9c-81df-3c94d124ab7e.html\" target=\"_blank\" rel=\"noreferrer noopener\">seeking to do just that<\/a>.&nbsp;<\/p>\n\n\n\n<p>But in recent weeks \u2014 in the wake of San Francisco-based Anthropic warning that its newest model, Mythos,&nbsp;<a href=\"https:\/\/fortune.com\/2026\/03\/26\/anthropic-says-testing-mythos-powerful-new-ai-model-after-data-leak-reveals-its-existence-step-change-in-capabilities\/\" target=\"_blank\" rel=\"noreferrer noopener\">represents a dangerous cybersecurity threat<\/a>&nbsp;\u2014&nbsp;Trump\u2019s administration seems to have embraced the need to control the technology.<\/p>\n\n\n\n<p>The administration\u2019s moves \u2014 including what amounts to a ban on Anthropic\u2019s Mythos and a preclearance regime for OpenAI \u2014 go well beyond anything the Biden administration put in place.&nbsp;<\/p>\n\n\n\n<p>\u201cIt\u2019s certainly a landmark about-face in the attitude that the [Trump] administration has about AI,\u201d said Anthony Aguirre, CEO of the Future of Life Institute, a nonprofit focusing on trying to prevent AI from causing large-scale harm.<\/p>\n\n\n\n<p>The Mythos ban in particular, he said, is a \u201c180-degree shift in how they\u2019re treating AI.\u201d<\/p>\n\n\n\n<p>There have been growing calls in recent years to regulate AI. Members of the public, advocacy groups, legal experts, politicians and even people within the industry have pushed for laws or rules that would prevent AI models from doing harmful things such as&nbsp;<a href=\"https:\/\/www.sfexaminer.com\/news\/technology\/california-passes-bills-to-protect-kids-from-chatgpt-harms\/article_cc26f399-ce07-45ab-9c71-51f38f2de7d4.html\" target=\"_blank\" rel=\"noreferrer noopener\">encouraging teen suicides<\/a>,&nbsp;<a href=\"https:\/\/www.sfexaminer.com\/news\/technology\/watchdogs-say-grok-deepfakes-still-incredibly-easy-to-make\/article_d9ae4bca-91e3-4169-b819-c58fcd4671b9.html\" target=\"_blank\" rel=\"noreferrer noopener\">creating nonconsensual sexual deepfakes<\/a>, exploiting computer vulnerabilities, or&nbsp;<a href=\"https:\/\/www.sfexaminer.com\/news\/technology\/california-legislature-passes-contentious-wiener-ai-bill\/article_d171ec96-659c-11ef-8761-b7e65ff2895a.html\" target=\"_blank\" rel=\"noreferrer noopener\">creating bioweapons or other weapons of mass destruction<\/a>.&nbsp;<\/p>\n\n\n\n<p>But regulatory and security experts are dubious about the rationale, legality and effectiveness of the administration\u2019s recent moves to constrain the technology, particularly the ban on Mythos.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote\">\n<p>\u201cMy problem is not so much that they\u2019ve changed their minds,\u201d said Alan Rozenshtein, a law professor at the University of Minnesota law school, \u201cit\u2019s just that I have no confidence in the policymaking apparatus of this administration.&#8221;<\/p>\n<\/blockquote>\n\n\n\n<p>While the administration might have been acting out of real concern about the model\u2019s cybersecurity implications, it has given little public justification for its move, there doesn\u2019t seem to be a coherent process undergirding it, and there\u2019s reason to believe it might have been driven as much by animosity toward Anthropic as anything else, those experts told The Examiner.<\/p>\n\n\n\n<p>Because of the importance of AI technology and its implications on society and the economy, it\u2019s no surprise that the Trump administration has changed its mind about regulating it, said Alan Rozenshtein, a law professor at the University of Minnesota law school who focuses on technology policy.<\/p>\n\n\n\n<p>\u201cMy problem is not so much that they\u2019ve changed their minds,\u201d Rozenshtein said. \u201cIt\u2019s just that I have no confidence in the policymaking apparatus of this administration.&#8221;<\/p>\n\n\n\n<p>Anthropic rolled out a preview of Mythos in early April. Warning that the model was&nbsp;<a href=\"https:\/\/www.anthropic.com\/research\/mythos-preview\" target=\"_blank\" rel=\"noreferrer noopener\">particularly good at finding and taking advantage of previously unknown computer-software vulnerabilities<\/a>, the company initially limited access to a handful of companies and organizations to use for defensive purposes.<\/p>\n\n\n\n<p>The release of the model reportedly set off alarm bells in the Trump administration, which quickly reversed its previous deregulatory stance.&nbsp; &nbsp;<\/p>\n\n\n\n<p>First, the president issued an executive order&nbsp;on June 2 calling for a process by which AI developers&nbsp;<a href=\"https:\/\/www.whitehouse.gov\/presidential-actions\/2026\/06\/promoting-advanced-artificial-intelligence-innovation-and-security\/\" target=\"_blank\" rel=\"noreferrer noopener\">could voluntarily submit their models for evaluation by the federal government<\/a>.&nbsp;<\/p>\n\n\n\n<p>Then, on June 9, administration officials&nbsp;<a href=\"https:\/\/www.theinformation.com\/articles\/trump-administration-asks-openai-stagger-release-new-model-security-concerns?rc=clhzk8\" target=\"_blank\" rel=\"noreferrer noopener\">held a secret meeting with AI companies<\/a>&nbsp;\u2014 pointedly not including Anthropic \u2014 about developing a process for reviewing cutting-edge AI models before they\u2019ve been released, according to The Information.<\/p>\n\n\n\n<p>That same day, Anthropic&nbsp;<a href=\"https:\/\/www.anthropic.com\/news\/claude-fable-5-mythos-5\" target=\"_blank\" rel=\"noreferrer noopener\">publicly released a version of Mythos<\/a>&nbsp;called Fable 5 to which it had added safeguards. It made available to select organizations a more capable version called Mythos 5.<\/p>\n\n\n\n<p>Three days later, the administration&nbsp;<a href=\"https:\/\/www.bloomberg.com\/news\/articles\/2026-06-16\/read-the-lutnick-letter-that-led-anthropic-to-disable-mythos\" target=\"_blank\" rel=\"noreferrer noopener\">put an export control on both models<\/a>&nbsp;that&nbsp;<a href=\"https:\/\/www.anthropic.com\/news\/fable-mythos-access\" target=\"_blank\" rel=\"noreferrer noopener\">effectively blocks the company from allowing anyone to use either one<\/a>.<\/p>\n\n\n\n<p>Sam Altman, CEO of Anthropic\u2019s cross-town rival OpenAI, told employees Wednesday the company would initially provide limited access to its newest model, GPT 5.6, in response to a request from the administration, The Information reported.<\/p>\n\n\n\n<p>Commerce Secretary Howard Lutnick had warned Altman earlier in the week to get sign-off from other government agencies before releasing the model, according to the report.<\/p>\n\n\n\n<p>The reason many regulatory and security experts are doubtful about the rationale for the Mythos ban is that it came mere months after Anthropic&nbsp;<a href=\"https:\/\/www.sfexaminer.com\/news\/technology\/anthropic-dispute-points-to-holes-in-ai-governance\/article_6c422372-81e7-4b1e-a99b-da118f357883.html\" target=\"_blank\" rel=\"noreferrer noopener\">got into a dispute with the administration over the use of its models by the military<\/a>. The company tried to forbid its technology from being used for domestic surveillance or fully autonomous weapons.<\/p>\n\n\n\n<p>In reaction to that dispute, Trump&nbsp;<a href=\"https:\/\/www.yahoo.com\/news\/articles\/anthropic-says-latest-pentagon-contract-195627154.html\" target=\"_blank\" rel=\"noreferrer noopener\">ordered the federal government to cut ties with the company<\/a>. Defense Secretary Pete Hegseth&nbsp;<a href=\"https:\/\/www.nytimes.com\/2026\/02\/27\/us\/politics\/anthropic-military-ai.html\" target=\"_blank\" rel=\"noreferrer noopener\">designated it a supply-chain risk<\/a>, threatening its relationships with other government contractors.<\/p>\n\n\n\n<p>Although a Pentagon official said in May the administration\u2019s&nbsp;<a href=\"https:\/\/www.cnbc.com\/2026\/05\/01\/pentagon-anthropic-blacklist-mythos-michael.html\" target=\"_blank\" rel=\"noreferrer noopener\">concern about Mythos is a \u201cseparate\u201d issue<\/a>, many experts doubt that\u2019s true.<\/p>\n\n\n\n<p>Soon after Anthropic released its Mythos preview and announced the model had found a collection of previously undiscovered vulnerabilities, security researchers elsewhere said they\u2019d found that other models that were already available, including OpenAI\u2019s GPT 5.5,&nbsp;<a href=\"https:\/\/www.theguardian.com\/commentisfree\/2026\/may\/08\/how-dangerous-is-anthropics-mythos-ai\" target=\"_blank\" rel=\"noreferrer noopener\">had similar or even greater capabilities<\/a>.<\/p>\n\n\n\n<p>But the administration didn\u2019t seek to block those models \u2014 although it did later seek to limit the release of more advanced OpenAI models.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Ex \/\/ Top Stories<\/h3>\n\n\n\n<h4 class=\"wp-block-heading\"><a href=\"https:\/\/www.sfexaminer.com\/culture\/visual-arts\/fort-mason-exhibit-shows-us-flags-stained-by-british-artist\/article_3db28364-d84f-4e83-a23b-0e81da00b0a7.html#tncms-source=top-stories-article\">\u2018The land remembers\u2019: Goldsworthy exhibit timed with America&#8217;s 250th<\/a><\/h4>\n\n\n\n<p>San Francisco&#8217;s waterfront will host &#8220;Red Flags,&#8221; a monumental exhibition making its West Coast debut.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><a href=\"https:\/\/www.sfexaminer.com\/news\/community\/sf-brings-bay-area-fifa-world-cup-excitement-to-residents\/article_f0b73a22-ce99-454d-b326-5136610ce513.html#tncms-source=top-stories-article\">No stadium, no problem. SF still has FIFA World Cup excitement<\/a><\/h4>\n\n\n\n<p>Though matches are taking place 45 miles away from The City, neighborhoods have been rocking with large gatherings<\/p>\n\n\n\n<h4 class=\"wp-block-heading\"><a href=\"https:\/\/www.sfexaminer.com\/news\/transit\/golden-gate-bridge-tolls-to-increase-with-bus-ferry-fares\/article_9c8fd8c8-c6df-4669-840e-119d97b718b0.html#tncms-source=top-stories-article\">Drivers can expect to pay more when crossing Golden Gate Bridge<\/a><\/h4>\n\n\n\n<p>A 50-cent hike is going into effect as part of a five-year toll program first approved in March 2024 to help address a financial shortfall<\/p>\n\n\n\n<p>\u201cPretty sure this government is mad at Anthropic and this is, like, the hammer they found,\u201d said Bruce Schneier, a renowned cybersecurity expert who is a lecturer at Harvard\u2019s Kennedy School. \u201cI don\u2019t think we can read any policy into the administration.\u201d<\/p>\n\n\n\n<blockquote class=\"wp-block-quote\">\n<p>\u201cPretty sure this government is mad at Anthropic and this is, like, the hammer they found,\u201d said Bruce Schneier, a cybersecurity expert and lecturer at the Harvard Kennedy School.<\/p>\n<\/blockquote>\n\n\n\n<p>Experts also questioned the legality of the export controls the administration slapped on the Mythos models.<\/p>\n\n\n\n<p>In his letter to Anthropic announcing the export controls, Lutnick&nbsp;<a href=\"https:\/\/www.bloomberg.com\/news\/articles\/2026-06-16\/read-the-lutnick-letter-that-led-anthropic-to-disable-mythos\" target=\"_blank\" rel=\"noreferrer noopener\">cited a collection of laws and regulations<\/a>&nbsp;seemingly giving him the authority to take that step. But some of the rules he cited appear irrelevant in this case, and he declined to name other regulations that might have given him clearer authority in the matter, said Bahrad Sokhansanj, a senior research scholar at the Institute for Law &amp; AI, a nonpartisan think tank.&nbsp;<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" loading=\"lazy\" width=\"4\" height=\"3\" src=\"https:\/\/occupysf.net\/wp-content\/uploads\/2026\/07\/image-3.png\" alt=\"\" class=\"wp-image-48975\"\/><figcaption class=\"wp-element-caption\">Commerce Secretary Howard Lutnick, right and President Donald Trump \u2014 seen in the Oval Office on June 11 \u2014 have both played roles in the White House\u2019s dustup with Anthropic.Jacquelyn Martin\/AP Photo<\/figcaption><\/figure>\n\n\n\n<p>One big objection to the use of an export control to block Mythos is that there\u2019s good reason to think that such measures&nbsp;aren\u2019t applicable in this case, the experts said.&nbsp;<\/p>\n\n\n\n<p>For nearly two decades, the Bureau of Industry and Security, which oversees such controls, has taken the position that&nbsp;<a href=\"https:\/\/natlawreview.com\/article\/exporting-cloud-decision-us-bureau-industry-and-security-bis\" target=\"_blank\" rel=\"noreferrer noopener\">digital services accessed in the cloud don\u2019t involve exporting<\/a>&nbsp;the underlying software or technology, so access to them can\u2019t be blocked with export controls.&nbsp;<\/p>\n\n\n\n<p>Mythos and other closed-source artificial-intelligence models are essentially cloud-based services. Consumers interact with them via web interfaces or apps. Businesses tap into them using application programming interfaces. Data is sent to the models via those APIs, and the models return responses, but the underlying technology is never transferred.&nbsp;<\/p>\n\n\n\n<p>There\u2019s a bill in Congress now that would extend export controls to cover APIs, suggesting that Congress doesn\u2019t believe the BIS currently has that power, Rozenshtein said.<\/p>\n\n\n\n<p>\u201cIt\u2019s not clear even under what authority this export control is being issued,\u201d he said.<\/p>\n\n\n\n<p>Additionally, the regulations cited by Lutnick typically require investigations and detailed findings, which the administration seems to have skipped over entirely, Sokhansanj said. Also, before putting in place severe measures such as the one they took, officials are supposed to explore whether there are less-restrictive options that would involve less harm to American companies, he said.<\/p>\n\n\n\n<p>\u201cThere\u2019s no indication they went through any of this kind of analysis, and there\u2019s no way they could have, because this stuff takes time,\u201d Sokhansanj said.<\/p>\n\n\n\n<p>But even if the move to slap export controls on the Mythos models was warranted and legal, at least some of the experts doubted their effectiveness. Other models that appear to be just as capable at finding and exploiting computer vulnerabilities as the Mythos ones are already available, the experts said.<\/p>\n\n\n\n<p>Even if the Mythos models have some capabilities that go beyond other current models, AI development in this area is proceeding so rapidly that other models will likely obtain those capabilities within months, they said.&nbsp;<\/p>\n\n\n\n<p>\u201cIf the goal is a safety one &#8230; I don\u2019t know if it will actually be helpful,\u201d said Elizaeth Seger, a senior policy advisor at the Tony Blair Institute, a U.K.-based public-policy think tank, of the export controls.&nbsp;<\/p>\n\n\n\n<p>\u201cIt\u2019s really only a matter of time before we start seeing equivalent capabilities that are made openly available on global markets for anyone to use and build off of,\u201d she said.<\/p>\n\n\n\n<p>AI technology does pose real dangers, the experts said. The U.S. government should be taking them seriously and taking steps to protect the public, they said. But many of the experts said that doing so would likely involve Congress writing laws and a democratic process of figuring out the appropriate regulations, rather than the reactive, seemingly one-off process the Trump administration is following.<\/p>\n\n\n\n<blockquote class=\"wp-block-quote\">\n<p>AI technology does pose real dangers, the experts said. The U.S. government should be taking them seriously and taking steps to protect the public.<\/p>\n<\/blockquote>\n\n\n\n<p>A big part of the problem with the way the administration has gone about AI regulation is that there\u2019s no transparency, said Amos Toh, a senior counsel at the Brennan Center for Justice, a left-leaning policy think tank at New York University\u2019s law school. Ideally, there would be a publicly determined and disclosed process for testing new models, evaluating the risks they pose and deciding how to mitigate those risks, he said.<\/p>\n\n\n\n<p>\u201cAll of these questions need to be sorted out \u2014 and not behind closed doors,\u201d Toh said.<\/p>\n\n\n\n<p>But Congress hasn\u2019t stepped forward to create that process. And the administration has been gutting the very agencies that would be necessary for overseeing that process, the experts said,&nbsp;<a href=\"https:\/\/techcrunch.com\/2026\/02\/25\/us-cybersecurity-agency-cisa-reportedly-in-dire-shape-amid-trump-cuts-and-layoffs\/\" target=\"_blank\" rel=\"noreferrer noopener\">most notably the Cybersecurity and Infrastructure Security Agency<\/a>.&nbsp; &nbsp;<\/p>\n\n\n\n<p>The Trump administration now seems to have dropped its laissez-faire, pro-AI industry approach. Pressure from the public and increasing risks from AI models will likely push it and Congress to go farther down the road toward regulation, some of the experts said.<\/p>\n\n\n\n<p>The public has become very skeptical of AI technology and overwhelmingly supports regulating it, Aguirre said. The questions that remain are what form regulation will take, whether the process will remain ad hoc and, in the end, how effective it will be.<\/p>\n\n\n\n<p>\u201cI think ultimately, public concern about this is not going to go away,\u201d Aguirre said.<\/p>\n\n\n\n<p><em>If you have a tip about tech, startups or the venture industry, contact Troy Wolverton at&nbsp;<\/em><em><a href=\"mailto:twolverton@sfexaminer.com\">twolverton@sfexaminer.com<\/a>&nbsp;or via text or Signal at (415) 515-5594.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"<p>When it comes to artificial intelligence, President Donald Trump\u2019s second administration has taken a decidedly antiregulatory stance \u2014 until this month.&nbsp; One of Trump\u2019s first moves in his second term was to&nbsp;ditch the modest AI safety guidelines issued by his predecessor, Joe Biden. After Congress over the summer overwhelmingly&nbsp;rejected a&#8230; <a class=\"continue-reading-link\" href=\"https:\/\/occupysf.net\/index.php\/2026\/07\/01\/anthropic-ban-not-the-ai-curb-thats-needed-security-experts-say\/\"> Continue reading <span class=\"meta-nav\">&rarr; <\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2030],"tags":[2138,2139,335],"_links":{"self":[{"href":"https:\/\/occupysf.net\/index.php\/wp-json\/wp\/v2\/posts\/48972"}],"collection":[{"href":"https:\/\/occupysf.net\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/occupysf.net\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/occupysf.net\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/occupysf.net\/index.php\/wp-json\/wp\/v2\/comments?post=48972"}],"version-history":[{"count":1,"href":"https:\/\/occupysf.net\/index.php\/wp-json\/wp\/v2\/posts\/48972\/revisions"}],"predecessor-version":[{"id":48977,"href":"https:\/\/occupysf.net\/index.php\/wp-json\/wp\/v2\/posts\/48972\/revisions\/48977"}],"wp:attachment":[{"href":"https:\/\/occupysf.net\/index.php\/wp-json\/wp\/v2\/media?parent=48972"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/occupysf.net\/index.php\/wp-json\/wp\/v2\/categories?post=48972"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/occupysf.net\/index.php\/wp-json\/wp\/v2\/tags?post=48972"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}