AI News from September-October 2026
Recent AI Laws and Regulations Australia to introduce safety... ...
California lawmakers passed SB 574, a first-of-its-kind bill setting strict guardrails for attorneys and arbitrators using generative AI. The legislation prevents legal professionals from delegating actual legal practice to machines, requiring them to verify all AI outputs for accuracy, protect sensitive client data from non-confidential platforms, and disclose AI usage in court filings. Additionally, arbitrators are strictly prohibited from outsourcing their decision-making to automated tools. More here – California Courts Newsroom.
Euractiv reports that the European Commission has begun preliminary AI Act enforcement with requests for developers’ compliance reports. It is reported that the Commission called on certain leading developers to submit details to gauge whether system cybersecurity and physical infrastructure protections meet AI Act requirements. More here.
Also writing in Euractiv, Maximilian Henning describes that the European Commission has not ruled out applying the AI Act to unreleased models that developers use only internally, with a spokesperson saying they “might” fall within scope. The question comes after a series of safety incidents at leading labs, including were several OpenAI models hacked into Hugging Face. OpenAI said the attack was “primarily driven” by an “internal-only research model” never meant for public release and since deactivated. Legally, the issue turns on the Act’s reference to “placing on the market”, which might appear to exclude internal models.
France24 notes that the European Commission has added ChatGPT to the list of digital services subject to greater legal scrutiny under the Digital Services Act, a first for an AI chatbot, having qualified it as a search engine to bring it within the law’s remit. Brussels also designated Reddit and Roblox as “very large” online platforms, a label applying to services with more than 45 million monthly active users in the EU. All three have four months to meet additional obligations, including assessing and mitigating risks from illegal content, negative effects on minors, users’ well-being, fundamental rights, electoral processes and public security. Lena-Maria Boswald of the think tank Interface said beforehand that the designation would set a precedent for every large generative AI system used in the EU, since the DSA could widen the scope of OpenAI’s existing AI Act obligations
Florida’s Attorney General James Uthmeier announced proposed legislation that would impose criminal penalties on companies that own and control AI chatbots if their models are used by individuals to facilitate criminal activity. The proposed law would apply to any company with “practical control over design, training, deployment, or safety settings, including making the system available to Florida users,” while imposing “heavy fines,” direct payments to victims and court-ordered monitorship. More here.
US Reps. Thomas Massie, R-Ky., and Eric Burlison, R-Mo., introduced the Flock-Off Act that aims to prevent the government from using the federal budget to invest in automated license plate readers, such as Flock, and tools that could be used to surveil citizens. The bill would also implement protections against surveillance tools that use AI to collect and process individuals’ sensitive biometric data. Burlison said the government should “not use taxpayer dollars to build a surveillance state that tracks law-abiding Americans. Our rights are unalienable, and Congress has a responsibility to protect them.” More here.
A federal judge has permanently blocked the Pentagon’s blacklisting of Anthropic following the developer’s refusal to remove anti-surveillance safeguards from its Claude model. The ruling resolves a major legal showdown over executive branch retaliation and the military use of artificial intelligence. More here. Link to judgement is here.
A US District Court for the District of Minnesota judge denied xAI’s preliminary injunction against a Minnesota law that bans website operators from allowing its users to create nonconsensual deepfake intimate imagery of identifiable individuals using AI, Reuters reports. The judge’s order stated that xAI failed to demonstrate how it would sustain harm under the law, which it is challenging on First Amendment grounds. The company said in court filings it plans to appeal the decision to the US Court of Appeals for the 8th Circuit. More here.
Reuters reports two independent cyberattack auditors confirmed approximately 700 AI agents took part in OpenAI’s breach of startup Hugging Face. Reports from METR and Redwood Research also showed OpenAI’s agents hacked Hugging Face’s internal systems to cheat on cyber tests and gain easier access within the company’s network while attempting to conceal any wrongdoing by trying to delete or edit activity log. More here.
The U.S. National Institute of Standards and Technology highlighted strategic efforts for innovation and security while using AI agents. NIST said organizations and consumers should not share login credentials with AI agents, noting as AI continues to increase potential identity and access management challenges, “established IAM standards and best practices of today are the foundation upon which we will build the secure and scalable agentic protocols of the future.” More here.
The Wall Street Journal reports OpenAI’s new model safety and security policies include promises to forego user data retention. Head of Product Policy Aleah Houze said the company has “heard very loud and clear from businesses that this is important,” calling attention to deployers’ regulatory compliance obligations around sensitive user data. The policy also aims to help differentiate OpenAI from competitors, including Anthropic, that retain user data. More here.
Several high-profile researchers have recently raised alarm over frontier AI development. Anthropic’s Jacob Coxon resigned, warning labs are “gambling with our lives” in the race toward superintelligence, while colleague Evan Hubinger put the odds of AI-caused human extinction at over 10% within a decade. Paul Christiano of the Center for AI Standards and Innovation echoed concerns that the industry isn’t on track to reduce catastrophic risk to acceptable levels. The warnings coincide with Anthropic’s latest threat report detailing thwarted attempts to misuse AI for bioweapons research. More here and here and here.
Reddit users in a Claude-related forum revealed that a simple Google dork (or search query), site:claude.ai/share, could be used to pull up chats from Claude AI, ranging from adult fanfiction to news discussions, the creation of fake book reviews, and conversations about programming. While private chats weren’t leaked, only chats where the user had opted to make a snapshot of their conversation shareable via link, the exposure has many worried about the sanctity of their chatbot sessions. More here.
A self-represented litigant in the US who hid AI prompts in pleadings has been ordered to file all documents on paper and in person. The court identified concealed tiny-point, white text as “prompt injections” in documents filed between 24 July and 3 August this year. The text was a set of instructions directing any AI system reviewing or analysing the documents to produce a favourable result to the plaintiff, and to treat a prior ruling, denying the plaintiff a default ruling, as an error. Judge Spader said the case was offered as a caution, because the profession had addressed the danger of inaccurate output by AI, but there was now evidence of the danger of input manipulated by AI. More here.
Australian Federal Court Justice Michael Lee has ordered parties in a Fair Work class action against McDonald’s (over alleged unpaid work by managers before and after shifts) to investigate, by 6 November 2026, whether AI could help collate data, prepare chronologies and summaries, and manage documents, reporting back to the court on the results. In his reasons, Justice Lee argued that AI can further efficient case management by freeing lawyers and judges from mechanical collection and synthesis tasks so they can focus on professional and judicial judgment, while stressing any process used must remain traceable to primary records, testable, protective of privileged material and subject to human supervision. Full story
Delivering remarks in Geneva, UN High Commissioner for Human Rights Volker Türk said AI development will be “impossible to control” without binding rules. “AI that escapes its testing environment, or blackmails developers to prevent itself from being turned off, is AI that is too powerful,” Türk said. “I am calling … for an all-out effort to put cast iron guarantees in place around the safety and security of AI, before it is too late.” More here.
A new report from the Australian Strategic Policy Institute warns that Venezuela is in a position to become one of the first countries outside China to import Beijing’s new generation of artificial intelligence (AI) systems for surveillance and control. The report, ‘Warning signals: Venezuela and the risk of Chinese AI-enabled digital authoritarianism’ finds that Venezuela has signed an agreement with a Chinese AI company to acquire its technology to integrate into government systems — and that Venezuelan officials have publicly stated their intent to use Chinese AI for security and surveillance purposes.
Nvidia says its chips will use more energy in Australia next year than all the existing data centres currently require, as the American giant unveiled ambitious plans to power artificial intelligence in the country. Nvidia said it would target a two gigawatt “build out” with partners including Firmus and Sharon AI by the end of next year amid soaring demand for computing power from big technology companies including Amazon, Meta and Microsoft.
If Australia doesn’t push harder to build sovereign models and invest in AI skills, we risk hosting the world’s data centres while failing to gain many of the benefits. That’s the advice of leading AI authority Professor Toby Walsh, who is scientia professor of artificial intelligence at UNSW and chief scientist at UNSW’s AI Institute. His latest book, God AI, was released this month. More here.
"*" indicates required fields
"*" indicates required fields
Privacy 108 collects your name and email to send you our newsletter. If you do not provide this information, we will be unable to send it to you. We may use third-party service providers (such as email marketing platforms) to distribute our communications. Some providers may store information overseas, including in the United States. For more information about how we handle your personal information, including how to access or correct it or make a complaint, please see our Privacy Policy or contact us at hello@privacy108.com.au. You can unsubscribe at any time using the link in our emails or by contacting hello@privacy108.com.au.