{"id":"camfeed-20","numeric_id":20,"short_title":"Australian agencies, businesses seek access to 'dangerous' AI model","seo_title":"AI firm Anthropic's 'dangerous' Mythos model fuelling government's desire to lure company to Australia - ABC News","social_title":"Banks, telcos briefed on threat of 'dangerous' AI model","headline":"AI firm Anthropic's 'dangerous' Mythos model fuelling government's desire to lure company to Australia","teaser_title":"Banks, telcos briefed on threat of 'dangerous' AI model","titles_source_url":"https://www.abc.net.au/news/2026-05-20/anthropic-jeffery-bleich-in-australia-for-government-talks/106696532","titles_fetched_at":"2026-09-05T04:20:25+00:00","title":"Banks, telcos briefed on threat of 'dangerous' AI model","canonical_url":"https://www.abc.net.au/news/2026-05-20/anthropic-jeffery-bleich-in-australia-for-government-talks/106696532","published_at":"2026-05-20T06:41:55+00:00","source":"ABC News","source_type":"abc","type":"news article","content_role":"primary","author":"Cameron Wilson","summary":"A new generation of AI models deemed \"too dangerous\" for public release is driving Australia's efforts to lure top AI companies to establish a major presence in the country.","text":"A new generation of AI models deemed \"too dangerous\" for public release is driving Australia's efforts to lure top AI companies to establish a major presence in the country.","has_full_text":true,"mcp_visible":true,"mcp_full_text_allowed":true,"full_text_word_count":1196,"full_text_status":"ok","full_text_source_url":"https://www.abc.net.au/news/2026-05-20/anthropic-jeffery-bleich-in-australia-for-government-talks/106696532","full_text_fetched_at":"2026-08-05T10:44:53+00:00","has_transcript":false,"transcript_status":"","transcript_kind":"","transcript_char_count":0,"topics":["AI","technology policy","platforms"],"quoted_author":"","quoted_url":"","quoted_text":"","citation":"Cameron Wilson, \"Banks, telcos briefed on threat of 'dangerous' AI model\", ABC News, 2026-05-20, https://www.abc.net.au/news/2026-05-20/anthropic-jeffery-bleich-in-australia-for-government-talks/106696532","citation_markdown":"Cameron Wilson, [Banks, telcos briefed on threat of 'dangerous' AI model](https://www.abc.net.au/news/2026-05-20/anthropic-jeffery-bleich-in-australia-for-government-talks/106696532) — ABC News — 2026-05-20","attribution":"Originally published at ABC News. Link to the canonical ABC URL: https://www.abc.net.au/news/2026-05-20/anthropic-jeffery-bleich-in-australia-for-government-talks/106696532","related_items":[{"id":"camfeed-28","title":"New AI models 'too dangerous' to release are fuelling the government's desire to woo tech companies down under","canonical_url":"https://webcam.substack.com/p/new-ai-models-too-dangerous-to-release","public_data_url":"https://camfeed.cameronwilson.com.au/items/28.json","source_type":"substack","source":"webcam","type":"newsletter commentary","content_role":"commentary","published_at":"2026-05-20T11:34:01+00:00","relation":"Newsletter commentary on ABC article: Banks, telcos briefed on threat of 'dangerous' AI model","excerpt":"G’day! I hope you don’t mind but I’m going to start using this list more regularly to send out my reporting (which, now in my new gig as the national AI reporter at the ABC, is no longer behind a paywall). Just in case, here’s an unsubscribe button if you’re not keen! unsubscrib...","has_transcript":false,"transcript_status":"","transcript_kind":""}],"full_text":"A new generation of AI models deemed \"too dangerous\" for public release is driving Australia's efforts to lure top AI companies to establish a major presence in the country.\n\nThe chief lawyer of AI company Anthropic, Jeffrey Bleich, flew into Australia this week for meetings with senior government officials.\n\nOn the agenda for the meetings is obtaining access to Anthropic's newest model, Claude Mythos.\n\nMythos and a similar new model from rival OpenAI are being touted as so capable at finding cybersecurity weaknesses that the companies fear major disruptions if the technology falls into the wrong hands.\n\nNow, Australian agencies and companies are pushing for access to the models, adding to the buzzy AI company's leverage as it considers where it will set up major hubs outside the US.\n\nA federal government briefing, obtained under freedom of information laws, shows Australia wants \"major investments in Australia\" and is willing to discuss reviewing copyright laws as part of a deal.\n\nAnthropic briefs government, critical infrastructure on new AI cyber risks\n\nMr Bleich, who served as Barack Obama's ambassador to Australia from 2009 to 2013 and is now Anthropic's general counsel, met with Assistant Minister for Science, Technology and the Digital Economy, Andrew Charlton, staff from the Department of Home Affairs and the Australian Signals Directorate.\n\nA major change since the company's US executives visited in April is the limited release of Mythos.\n\nWhile AI makers often hype the capabilities of their products, organisations such as UK's AI Safety Institute and Mozilla, maker of web browser Firefox, have tested and vouch for the advanced capabilities of Mythos when it comes to finding bugs that could be exploited for cyber attacks.\n\nPreviously, Anthropic had released its models to any paying subscriber. But this time, the company chose to limit access to Mythos to about 50 mostly US-based technology companies such as Amazon, Microsoft, Apple and Google.\n\nNo Australian companies or organisations were on the list.\n\nOpenAI has followed suit and is only releasing its newest cybersecurity-based model, GPT-5.5-Cyber, to select customers.\n\nThe federal government declined to confirm or deny that its agencies and Australian companies still do not have access to Mythos.\n\nA spokesperson for the Department of Home Affairs said it had \"extensive discussions with Anthropic on the cyber security implications of AI and improved cyber defence mechanisms\".\n\nHome Affairs has co-ordinated multiple meetings between Anthropic and Australian operators of critical infrastructure since the release of Mythos.\n\nLate last month Anthropic held a briefing with 170 representatives from Australian finance, communications, transport, energy, data, food and grocery sectors.\n\nThese organisations are declared as Systems of National Significance, a subset of critical infrastructure providers considered crucial to national security.\n\nA source with knowledge of the closed-door meetings between the government and Anthropic but not authorised to speak publicly said that Australian access to Mythos was a topic of discussion.\n\nA spokesperson for OpenAI told the ABC it was starting to roll out its Mythos-like model, GPT-5.5-Cyber, to a limited number of Australian organisations.\n\n\"We have begun rollout of our Trusted Access for Cyber (TAC) program, including GPT-5.5-Cyber, to a handful of customers in Australia,\" they said. This includes Commonwealth Bank and Westpac, the Australian Financial Review reported.\n\nA spokesperson for Home Affairs minister Tony Burke said that the government was working with the private sector to prepare for developments in AI.\n\n\"Our government has put in place systems to help our country adapt to and resist evolving cyber threats,\" they said.\n\nAnthropic's spokesperson declined to comment on discussions in its meetings with government.\n\nAustralia's attempts to lure AI companies\n\nIn December the federal government released a national AI plan that included attracting big tech companies to invest here and mitigating risks from the technology.\n\nAI companies such as Anthropic and OpenAI, as well as more established players such as Google, Meta and Amazon, see Australia as an attractive destination for building AI infrastructure in the southern hemisphere. Australia's skilled workforce, energy grid and political stability are all drawcards.\n\nIn the United States, Anthropic faces immense pressure from the Trump administration over its decision to limit the use of its models by the US military. The company is reportedly considering major expansion overseas.\n\nThe federal government wants Anthropic to pick Australia. It stepped up its courtship with a meeting on the sidelines of an international AI summit in India between Anthropic CEO Dario Amodei and Charlton in February this year.\n\nDepartment of Industry briefing notes obtained through a freedom of information request reveal the government's thinking ahead of the meeting.\n\nUnder the heading \"What we want\", the document has three asks:\n\nThat Anthropic make \"major investments in ways that create benefits for Australia, including being additive to the energy grid\"\n\nThat it match commitments it made in the United States to cover grid infrastructure costs, build new power generation, reduce strain on the grid and invest in local communities\n\nThat it commit to working with Australia on frontier AI safety and security\n\nIt also seeks more information on Anthropic's major hurdles: \"[We want to] understand Anthropic's issues with Australia's copyright regime, how it affects their business and what alternatives have been explored.\"\n\nA spokesperson for the department of industry declined to comment on the briefing document.\n\nGroups representing tech companies and the private sector are arguing that Australia's copyright law limits AI investment in Australia.\n\nSo far, the Albanese government has ruled out giving AI companies an \"text and data mining\" exemption to train AI on Australian copyright material without payment.\n\nCreatives fear the government will agree to water down protections to clear way for investment for AI companies.\n\nSince the meeting in India the federal government and Anthropic have sought to formalise this increasingly close relationship.\n\nSix weeks later, Industry Minister Tim Ayres met with Anthropic CEO Dario Amodei in Australia and signed a memorandum of understanding outlining a \"shared vision\" with the company.\n\nThe memo contains several of the asks from the briefing in non-binding forms.\n\nThe company also opened its Sydney office last month, its fourth office in the Asia-Pacific.\n\n'Winners are being picked'\n\nCyberCX chief strategy officer and former Australian national cybersecurity advisor Alastair MacGibbon said that securing access to cutting edge AI tools for Australian firms and agencies isn't the be-all-and-end-all.\n\nA lot of Australia's digital infrastructure is built on systems created by major technology firms and big international financial institutions that have been given early access to Anthropic's new model, he said.\n\nOn the other hand, many organisations have their own bespoke software that will not be supported by companies that have had access to these AI models, Mr MacGibbon said.\n\n\"Winners are being picked, and [the tech companies] chose in a US-centric way who the winners were,\" he said.\n\nMr MacGibbon said the responsibility for closing that gap should fall on governments, not on individual companies trying to negotiate access on their own.\n\nAn Anthropic spokesperson declined to say when Australian organisations outside the existing partners will get access to Mythos.\n\n\"We've been having constructive conversations with government, financial institutions, and enterprise customers globally; and plan to expand access safely in coordination with them in due course,\" the spokesperson said.","full_text_attribution":"Full text extracted from the ABC News article at https://www.abc.net.au/news/2026-05-20/anthropic-jeffery-bleich-in-australia-for-government-talks/106696532. Attribute to ABC News and link to the canonical URL."}