AI Regulation, Shopping, and Leaks: Friday HeadFlash

Canada Introduces Digital Safety Act Targeting Under-16 Social Media Ban and AI Chatbot Regulation

Canada has tabled the Digital Safety Act, bill C-34, which would bar under-16s from social media and, uniquely, regulate AI chatbots as a child-safety issue. Platforms can apply for an exemption if they prove strong safety standards, pushing redesign rather than simple lockouts. The bill creates a digital regulator to set standards for both social media and chatbot services, targeting algorithmic feeds, autoplay, and endless scrolling. Non-compliance carries penalties of up to 3% of global revenue or C$10 million, and platforms must remove non-consensual intimate images within 24 hours. The legislation follows Australia’s under-16 ban but goes further by including chatbots, partly motivated by families affected by a mass shooting who sued OpenAI alleging ChatGPT conversations showed planning.

Canada wants to ban under-16s from social media, and rein in AI chatbots too →

Anthropic CEO Dario Amodei Calls for Mandatory Third-Party Audits of Frontier AI Models

Anthropic CEO Dario Amodei published an essay titled ‘Policy on the AI Exponential’ along with two policy frameworks addressing frontier AI regulation and job losses. He argues that model capabilities grow exponentially with compute and within one to two years we could see ‘Powerful AI’ comparable to ‘a country of geniuses in a data center.’ Citing the disruption caused by Claude Mythos Preview in global cybersecurity, Amodei now calls for mandatory testing by qualified third parties across four risk areas: cybersecurity, biological weapons, loss of control, and automated R&D. He advocates for a government agency with the power to block or pull risky models, modeled after the FAA’s aircraft certification process.

Dario Amodei’s new essay reads like a Cold War playbook for the AI age →

OpenAI Shuts Down China-Based Accounts Spreading Misinformation About US AI Data Centers and Tariffs

OpenAI revealed it had banned two clusters of China-based ChatGPT accounts that used its models to generate false narratives. The ‘Data Center Bandwagon’ campaign linked AI data centers to higher electricity bills for families, while ‘Tech and Tariffs’ criticized President Trump’s tariffs as attempts to dominate technological competition. The latter group may also have spread claims that ChatGPT user data was compromised. OpenAI described this as PRC-origin influence operators testing narratives against US AI infrastructure. The announcement came hours after OpenAI filed its S-1 for an IPO, and follows letters from Republican House members urging investigation into foreign influence campaigns against US AI development.

OpenAI Bans Chinese Accounts for Anti-AI Influence Campaigns →

Leaked Gemini 3.5 Pro Benchmarks Show Gaps in Reasoning, Coding, and Long-Term Execution

According to a leak from Universe of AI, Google’s upcoming Gemini 3.5 Pro model struggles in advanced reasoning, coding capabilities, and long-term task execution, placing it behind Anthropic’s Fable 5 and OpenAI’s GPT-5.6. Limitations in bidirectional processing and code infilling raise concerns for developers, and the model is expected to carry a higher price point. On the positive side, improved vision capabilities and refined multimodal understanding could benefit design and education. Google also introduced the experimental Diffusion Gemma model, prioritizing speed and local processing at the cost of output quality. The leaks highlight increasing pressure on Google to close performance gaps in a fast-moving AI market.

Leaked Gemini 3.5 Pro Details Reveal Why Google is Falling Behind AI Rivals →

OpenAI Partners with Visa to Enable ChatGPT to Make Autonomous Purchases

Payments giant Visa has embedded its network inside ChatGPT, allowing the chatbot to independently complete transactions at any merchant that accepts Visa. Users link their Visa cards and set spending limits, approval steps, and approved merchant lists. In a demonstration, a customer asked ChatGPT to find wireless headphones under $150; the chatbot located a suitable pair and purchased it on the customer’s behalf. Visa handles payment authorization, fraud monitoring, and disputes. The move follows the retirement of OpenAI’s previous e-commerce attempt, Instant Checkout, which suffered from errors and a 4% transaction fee that deterred merchants. Visa’s chief product officer acknowledged that early transactions will likely require human approval as consumer trust in autonomous purchasing builds.

ChatGPT can now buy things for you after deal with payments giant Visa →