U.S. Orders Anthropic to Shut Down Claude Fable 5 and Mythos 5 Worldwide
++ EU releases a transparency code for AI-generated content and deepfakes; Germany plans an AI Safety Institute; Google sues an alleged China-linked AI scam network.. & more
Today’s highlights:
The U.S. government has ordered Anthropic to immediately disable access worldwide to its Claude Fable 5 and Claude Mythos 5 models, citing national security concerns under an export control action. Anthropic said it complied but argued the move was based on only verbal evidence of a limited jailbreak in Fable 5 and said the model’s key protections still remain in place. Mythos had been tightly restricted because of its strong vulnerability-finding abilities, while Fable 5 was released publicly with added safeguards and was seen as Anthropic’s most capable public model. The episode is a setback for the company and highlights how its own safety warnings around advanced AI may have drawn the government scrutiny now affecting its business.
At the School of Responsible AI (SoRAI), we help both individuals and organizations build practical, real-world AI literacy and Responsible AI capability through structured, engaging, and action-oriented programs. For individuals, this includes AI Literacy, globally relevant certification training such as AIGP, RAI, and AAIA, as well as career transition and advisory support for professionals moving into AI governance roles. For organizations, we offer customized enterprise AI literacy training, Responsible AI strategy and governance setup, and AI assurance support to help teams understand, operationalize, and validate AI responsibly. At the core of SoRAI is a progressive three-layer approach: first helping people understand AI, then build the right governance foundations, and finally validate readiness through assurance and audit-focused thinking. Want to learn more? Explore our AI Literacy programs, certification trainings, and career support offerings, or write to us for customized enterprise solutions.
⚖️ AI Ethics
EU Commission Releases Code of Practice on Transparency for AI-Generated Content and Deepfakes
The European Commission has published a new Code of Practice on transparency of AI-generated content, setting out rules for providers to mark and detect AI-made or manipulated media and for deployers to label deepfakes and AI-generated text. The code, released on 10 June 2026, also includes a set of icons that generative AI deployers may use to label such content. It is now undergoing an adequacy assessment by the Commission and the AI Board and will be backed by separate guidelines on the AI Act’s Article 50 transparency obligations. If approved, signatories will be able to use the code to show compliance across the EU, while companies choosing other methods will face individual checks by market surveillance authorities.
Germany Plans AI Safety Institute to Assess Risks, Capabilities and Security Impact of Advanced Models
Germany said it plans to set up an AI safety institute to assess the performance, capabilities, and risks of advanced AI models, with a focus on national security and cybersecurity. The move follows discussions in the country’s National Security Council about the cyber implications of increasingly powerful AI systems. Officials said the institute will study deployment risks, improve information-sharing with similar bodies abroad, and support international cooperation on AI governance and technical standards. The plan comes as Germany steps up cybersecurity efforts after reporting about 334,000 cybercrime cases last year, around two-thirds of them linked to foreign or unknown sources.
Italy Regulator Drops WhatsApp AI Probe Into Meta as EU Expands Competition Investigation
Italy’s competition watchdog, AGCM, has dropped its investigation into Meta Platforms over allegations that the company abused its dominant position by integrating its AI tool into WhatsApp. The probe was opened in July 2025 and focused on whether Meta’s actions in the messaging service raised competition concerns. AGCM said it closed the case because the European Commission has expanded its own investigation of the same issue to cover Italy as well. The move shifts oversight of the matter from the national regulator to the EU level.
xAI Sued After Firing Engineer Who Raised Grok Safety Concerns, Lawsuit Claims
A former xAI engineer has sued xAI and SpaceX in California, alleging he was fired after repeatedly raising concerns about Grok’s safety and bias risks. The complaint says he warned that the chatbot could encourage discrimination and help spread dangerous information, and it points to later Grok controversies, including hateful outputs and misuse involving nonconsensual sexual imagery on X. The lawsuit claims xAI failed to prioritize legally required safety measures and says a senior company leader retaliated against him for pushing stronger safeguards, while not blaming Elon Musk directly. xAI and SpaceX had not commented at the time of reporting, and the engineer is seeking damages and a court ruling that the companies acted unlawfully.
Cybersecurity Researchers Criticize Anthropic Fable Guardrails for Blocking Benign Security and Coding Requests
Anthropic has released Fable, a public but limited version of its cybersecurity-focused model Mythos, but early reaction from security researchers has been critical because the model’s guardrails appear to block even harmless cyber-related requests such as code reviews or reading blog posts. When flagged, Fable pauses and says its safety systems detected cybersecurity or biology topics, reflecting Anthropic’s effort to prevent misuse for malware or biological threats. The company had previously kept Mythos largely restricted through Project Glasswing before expanding access to more organizations in 15 countries. Some experts said the limits seem overly broad and possibly keyword-based, though they also noted Anthropic may loosen and refine the guardrails over time.
Google Sues Alleged Chinese AI-Powered Cybercrime Network Behind Scams Targeting Hundreds of Thousands of Victims
Google has sued an alleged Chinese cybercrime network known as Outsider Enterprise, accusing it of running a large AI-assisted phishing operation that impersonated Google and other brands through scam texts and fake websites. According to Google and the FBI, the group sent millions of fraudulent messages, used about 1 million phishing domains and thousands of fake sites, and financially harmed hundreds of thousands of victims, with broader losses tied to the operation estimated at up to $1.9 billion. Court filings say the network sold easy-to-use phishing software, relied on Telegram for coordination, and used AI tools to quickly build convincing scam pages designed to steal passwords, payment details, and multi-factor authentication codes. Google said it is working with telecom carriers and the FBI to disrupt the infrastructure, while seeking damages and a court order to shut the operation down.
KPMG Withdraws AI Usage Report After Apparent Hallucinations Trigger False Claims About Multiple Organizations
KPMG has withdrawn a report on agentic AI after several organizations said its claims about their use of AI were false or misleading. GPTZero told the Financial Times that it found multiple inaccuracies in the October 2025 report and said they appeared to be caused by AI hallucinations. UBS, the UK’s National Health Service, Swiss Federal Railways, and Transport for London all disputed how the report described their AI adoption. KPMG said it removed the report from its websites while it investigates and added that its staff are expected to follow rules on responsible AI use, including human review and source verification.
State Attorneys General Launch Investigation Into OpenAI Over Advertising, Data Practices, and User Safety
A coalition of state attorneys general has opened an investigation into OpenAI, with The Wall Street Journal reporting that New York’s attorney general served the company a subpoena seeking documents on advertising, user engagement, model behavior, and its handling of consumer, health, minor, and senior data. OpenAI said it takes the concerns seriously and will cooperate, adding that ChatGPT now includes stronger safeguards for minors and vulnerable users. The company has not disclosed which states are involved or the full scope of the request. The probe adds to OpenAI’s growing legal pressure, including lawsuits over copyright and product safety, and comes shortly after the company confidentially filed for an initial public offering.
Anthropic CEO Urges Tougher AI Rules as Emerging Risks Outpace Transparency Measures Worldwide
Anthropic’s chief executive said governments need tougher AI rules as advanced systems become more powerful and potentially dangerous, arguing that transparency alone is no longer enough. He called for mandatory third-party testing of frontier AI models above certain computing thresholds, with governments able to block deployment if systems pose unacceptable risks in areas such as cybersecurity, biological weapons, loss of control and automated research. He also warned that AI could cause major job displacement and said policymakers should track the impact, create employment incentives and consider long-term support such as taxes on AI beneficiaries, universal basic income or capital accounts. Beyond economics, he said regulators must adapt faster to AI-driven science, while democracies should tighten safeguards against AI-enabled surveillance, autonomous weapons and geopolitical imbalances.
🚀 AI Breakthroughs
Anthropic Opens Claude Fable 5 Public Access With Safety Limits and Enterprise Usage Controls
Anthropic has made Claude Fable 5, a public version of its more powerful Mythos model, available through its Claude API and consumption-based Enterprise plans, with temporary access on some subscriptions before shifting to usage credits. The company said Fable 5 is strong at coding, knowledge work, and vision, but it includes strict safety limits and will hand off certain high-risk requests in areas like cybersecurity, biology, and chemistry to Claude Opus 4.8. Anthropic is also giving approved organizations access to Mythos 5 while requiring 30-day traffic retention for both models to monitor jailbreaks and other novel attacks, even for some enterprise customers that previously had zero-retention terms. The model is priced at $10 per million input tokens and $50 per million output tokens, double Opus 4.8, as Anthropic expects strong demand despite growing enterprise concerns over AI costs.
Anthropic’s Claude Fable 5 Generates Playable Video Games and Complex Tools From a Single Prompt
Anthropic has released Claude Fable 5, described as the first public version of its closely watched Mythos model, and early testing suggests it can handle long, complex tasks with unusually strong performance. In reported demos, the model generated several playable video games from a single prompt, including simple arcade-style titles and more experimental literary experiences. It was also used to build an isochronic map, showing that its abilities extend beyond games into detailed visualization and software creation. The early takeaway is that AI tools are moving closer to producing projects that once needed larger teams, with faster progress in coding and interactive design.
Apple Details Siri AI, iOS 27, Apple Intelligence, and Search Updates at WWDC 2026
At WWDC 2026, Apple put the focus on catching up in AI and fixing long-standing software complaints, with a revamped Siri at the center of the event. The company said Siri now uses Google Gemini under the hood, making it more conversational, context-aware, and available both across apps and in a standalone app, while also extending Apple Intelligence features across Safari, Messages, Phone, Photos, Shortcuts, and search. Apple also detailed iOS 27 upgrades including broader device support back to the iPhone 11, performance improvements, new parental controls, systemwide AI dictation, and Health updates such as perimenopause tracking. Separately, references in the iOS 27 developer beta hinted at a possible foldable iPhone, though Apple did not confirm any such device during the keynote.
Amazon Adds AI-Powered Custom Merchandise Design in Shopping App, Expanding Challenge to Print-on-Demand Platforms
Amazon has added a new AI-powered tool in its Shopping app that lets U.S. customers create custom merchandise through Alexa prompts and order the results through Merch on Demand. Users can generate and edit designs for items such as T-shirts, hoodies, tumblers, and water bottles, while Amazon handles printing and Prime delivery. The feature is free to use, with customers paying only for the products they buy. The move brings AI-generated merch directly into Amazon’s retail app and increases pressure on print-on-demand platforms such as Redbubble, Bonfire, Spring, and Fourthwall.
Meta Signs First India AI Data Center Deal With Reliance for Jamnagar Facility
Meta has signed its first AI data center deal in India through a partnership with Reliance Industries, leasing capacity at a 168-megawatt AI-ready facility being built in Jamnagar, Gujarat. The site is expected to be operational within two years and will support Meta’s global AI and infrastructure needs, with Reliance handling construction, power, connectivity, and operations. The project will run on renewable energy and use desalinated seawater for cooling, while Meta will cover the energy and water costs tied to its operations. The agreement adds to a broader push by global and Indian companies to expand AI and cloud infrastructure in India, where data center capacity has grown sharply and is projected to rise further this decade.
Google Labs Tests AI App That Uses Gmail, Photos and Calendar Data for Personalised Daily Stories
Google Labs has unveiled Dreambeans, an experimental AI app that uses data from connected Google services such as Gmail, Calendar, Photos, YouTube and Search to create personalised daily stories for users. Instead of acting like a standard chatbot, the tool proactively surfaces recommendations, tips and information based on a person’s recent activity, upcoming plans and interests. Users can choose which services to connect, save stories for later and adjust personalisation settings, while each story also includes AI-generated illustrations. The release reflects a broader shift in consumer AI, with companies moving toward more context-aware assistants that automatically surface relevant content from a user’s digital life.
Google Rolls Out Gemini 3.5 Live Translate for Real-Time Speech Translation Across 70+ Languages
Google has released Gemini 3.5 Live Translate, a new speech-to-speech translation model designed to deliver fluid, natural voice translation in real time. The model can automatically detect more than 70 languages and generate translated speech that preserves a speaker’s tone, pacing, and pitch, while staying only a few seconds behind live speech. Unlike older turn-based systems, it translates continuously as people speak and is built to work in noisy environments without manual language setup. The rollout starts today for developers through the Gemini Live API and Google AI Studio, will reach enterprises in private preview in Google Meet this month, and is also coming to Google Translate on Android and iOS.
Lovable Data Study Finds Broader Participation in Software Creation and Emerging Economic Activity in 2026
Lovable’s 2026 data study says the gap between having a software idea and turning it into a working product has narrowed sharply over the past 18 months, opening software creation to more people beyond traditional developers. The report is based on Lovable’s product usage data from January 2025 to May 2026, along with a user survey conducted in May 2026. It presents early signs that real economic activity is starting to form around this shift, as more users turn ideas into working software. The study argues that the change is not only reshaping how software is built, but also expanding who can take part in the next wave of economic creation.
OpenAI Academy Adds Three New Courses Focused on AI Skills, Workflows, and Agents
OpenAI has added three new Academy courses aimed at helping companies build practical AI skills for the workplace: AI Foundations, Applied AI Foundations, and Agents and Workflows. The courses are designed to take employees from basic AI use, such as prompting and reviewing outputs, to creating repeatable workflows and managing agent-assisted tasks with human oversight. OpenAI said the program draws on input from its research, product, safety, and deployment teams, and is being supported with partners including BCG, Accenture, and BBVA. Each course offers a completion certificate, and the company said the broader Academy roadmap will continue to expand with updated content, reporting tools for organizations, and new role-based learning paths.
🎓AI Academia
RiskNet Dataset Tracks AI Risk Incidents From News With Alignment and Multidimensional Annotations
Researchers at Beijing University of Posts and Telecommunications have released RiskNet, a large-scale dataset designed to track real-world AI risk incidents reported in multilingual news coverage. The project aims to fill a gap in AI governance research, where existing incident databases are often small, manually curated, and difficult to use for continuous monitoring. RiskNet uses a structured process to identify AI risk news, screen event reports, align multiple stories about the same incident, and assign multi-dimensional labels to each case. According to the paper, the dataset turns scattered news reports into incident-centered records and also provides benchmark data for tasks such as event classification, incident matching, and risk labeling.
Synoptix Report Highlights Key Challenges in Operationalising JSP 936 for AI Assurance in UK Defence
A Synoptix report dated June 8, 2025 says the UK Defence AI assurance rulebook, JSP 936 Part 1, gives a strong governance starting point but remains difficult to put into practice. The review found 272 requirements in the directive and flagged about 35 as especially challenging, grouping them into eight problem areas including evidence standards, human interaction with AI, operational context, system integration, performance, safety and security, ethics, and AI complexity. The report says the main gap is not policy intent but the lack of mature methods, guidance, and organisational capability to assure AI in real-world defence settings. It concludes that safe and responsible military AI deployment will require iterative implementation, stronger assurance tools, and further support beyond the current directive.
Study Defines AI Legal Specialist as Independent Role Shaped by Expanding Global AI Regulation
A new academic paper argues that the fast-growing wave of AI regulation has created a need for a dedicated legal role focused specifically on AI governance, rather than relying on privacy lawyers, data protection officers, or compliance teams to stretch existing duties. It proposes an “AI Legal Specialist” as an independent legal professional trained to interpret and manage AI-related obligations under laws such as the EU AI Act, the Council of Europe’s AI convention, and policy frameworks emerging in the US, UK, Canada, Brazil, China, Japan, and Singapore. The paper says this role should stand on its own legal basis, not as an offshoot of technical standards or adjacent compliance jobs. It also outlines a skills framework linked to the European e-Competence Framework and suggests performance indicators that could help measure the role in practice, with the broader goal of supporting future international standardization, training, and adoption.
Systematic Review Finds AI Compliance Burdens and Regulatory Gaps Threaten Innovation in Critical Sectors
A new systematic literature review finds that AI adoption in critical sectors such as healthcare, finance, energy, and defense is advancing faster than the rules meant to govern it. Covering research and institutional sources published between 2020 and 2025, the paper identifies three main obstacles: fragmented regulations, heavy compliance costs that hit small and medium-sized firms especially hard, and governance models that do not match how modern AI systems are built and deployed. It warns that excessive regulation can slow valuable innovation, while weak oversight can allow biased, unsafe, or untested systems into high-stakes environments. The review says approaches such as risk-tiered regulation, compliance-by-design, and explainable AI could help policymakers and companies balance safety, trust, and innovation more effectively.
Study Warns Failed AI Systems Leave Residual Risks That Persist Beyond Decommissioning
A new paper argues that the risks of AI do not end when a system is shut down, warning that failed or withdrawn models can leave behind “AI debris” that continues to affect institutions. It says this residue can include broken accountability, biased data traces, staff deskilling, workflow dependence, and damage to trust that linger long after deployment ends. The study outlines how these effects persist through institutional habits, data feedback loops, and blame-shifting, and proposes an AI Debris Decommissioning Protocol to help auditors and regulators track cleanup after shutdown. Using Amazon’s scrapped hiring tool as an example, the paper says AI systems can keep shaping decisions even after they are officially removed.
About SoRAI: SoRAI is committed to advancing AI literacy through practical, accessible, and high-quality education. Our programs emphasize responsible AI use, equipping learners with the skills to anticipate and mitigate risks effectively. Our flagship AIGP certification courses, built on real-world experience, drive AI governance education with innovative, human-centric approaches, laying the foundation for quantifying AI governance literacy. Subscribe to our free newsletter to stay ahead of the AI Governance curve.
