AI Governance Is No Longer Optional: What Privacy Employers Are Really Asking For
If you’ve been reading our quarterly Privacy Job Repor... ...
The Council of the European Union adopted the proposed regulations of the Digital Omnibus on AI package. The package delays the effective dates for requirements on standalone high-risk models to 2 Dec. 2027, and 2 Aug. 2028 for high-risk models embedded in products. The framework also contains a new provision prohibiting any AI system that is capable of generating nonconsensual, deepfake intimate imagery or child sexual abuse material. Additionally, the AI Omnibus delays the enforcement data for member state authorities to establish regulatory sandboxes until 2 Aug. 2027 and addresses overlapping sectoral regulations that may impact AI development. More here.
The Illinois Artificial Intelligence Safety Measures Act received governor approval, increasing transparency and accountability requirements for AI models that generate more than US$500 m in annual revenue and includes a first-in-the nation requirement for annual independent third-party audits. The legislation, based on similar bills from California and New York, also adds new reporting requirements for potential large-scale harms caused by AI models. More here.
China has banned companies from offering virtual companion or virtual family-member services to minors, leading to companies like Alibaba and ByteDance (the owner of TikTok) to turn off their AI companion functionality. The law also requires companies “to build dedicated “minor modes” with usage-time limits, reminders to return to real-world interaction and enhanced parental controls”, reports AI news.
In May at London Court, an HR consultant used Garfield AI’s automated system to recover about £7000 she was owed by a hospitality business. The firm’s AI-powered litigation assistant generated the client’s pre-action correspondence, issued the claim, produced the documents, prepared the witness statements, then briefed a human barrister.1 After a three-hour contested hearing, the court found for her in full and dismissed the defendant’s counterclaim. More here.
Google has sued a Chinese phishing-as-a-service provider accused of teaching customers to use Gemini to generate and customize scam websites, a campaign linked to more than 1.59 million phishing URLs, over 100,000 victims, and widespread credential and financial theft. More here.
The White House announced it would pull back parts of its export ban on Anthropic’s Mythos 5 model after it previously restricted the company from releasing its frontier AI model due to cybersecurity concerns, Politico reports. Anthropic’s second frontier model, Fable 5, is still blocked. U.S. Secretary of Commerce Howard Lutnick said following the governments ban, Anthropic “has worked with the U.S. government to address risks associated with the Covered Models.” Meanwhile, Gov. Gavin Newsom, D-Calif., and Anthropic have come to an agreement over California’s use of their AI tools and will deploy the technology to state government. More here.
Anthropic’s Claude Fable 5 and Mythos 5 demonstrate major advances in coding and reasoning while raising new questions about exploit generation, enterprise AI spending, data governance and the growing urgency of faster vulnerability remediation. More here.
The White House is reportedly in “advanced” discussions with leading AI companies to establish a voluntary set of standards developers would adhere to before releasing new models, the Financial Times reports. The standards could be announced as early as 6 July and would create benchmarks for models that possess significant cyber capabilities. A source “familiar with the White House discussions” said the Center for AI Standards and Innovation and the National Security Agency would play an instrumental role in creating and monitoring the standards. More here (paywall).
The Five Eyes, the cybersecurity intelligence alliance, released a call to action last week on the use of frontier AI models, urging companies to bolster their security efforts due to the technology’s privacy implications. More here.
OpenAI will deploy its GPT-5.6 model 9 July after the U.S. government previously implemented restrictions to Anthropic and OpenAI’s frontier models due to concerns over the models’ advanced abilities, Axios reports. OpenAI conducted tests to ensure the tool’s safety with the National Institute of Standards and Technology’s Center for AI Standards and Innovation and plans to continue to test some of the tool’s functions before launching it to consumers. More here.
The Albanese government and global tech giants remain locked in a stalemate over law changes that would enable them to train their AI models in Australia. More here.
Meta paused its plan to use tracking technology to monitor employees and use the collected data to train its AI tools after the program faced employee pushback, the Guardian reports. More than 1,600 employees signed a petition urging the company to not use their computer data for AI training, noting collecting and sharing information, such as performance data, meeting transcripts and internal messages, “raises serious concerns around privacy, consent, and trust in the workplace.” More here.
The U.S. Federal Trade Commission is seeking public comment on its proposed policy statement that aims to ensure AI companies are not serving misleading or deceptive content to consumers. The policy looks to prevent AI systems from intentionally distributing potentially inaccurate or biased information in violation of Section 5 of the FTC Act. The comment deadline is 31 July. More here.
An independent panel of 40 leading scientists and experts will present their report on the benefits and risks of AI at the inaugural U.N. Global Dialogue on AI governance in Geneva 6-7 July, Reuters reports. Key findings from the preliminary report include the need for global policymakers to provide scientific evidence to govern AI, while at the same time, the technology’s capabilities are outpacing scientific understanding and governments’ ability to adapt to new advances. More here.
Using AI to draft court materials and contracts requires a high level of due diligence to catch errors and hallucinations. This isn’t a low-risk decision. 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.