dig.watch

OpenAI has explained the architecture behind GPT-Live, the real-time system powering ChatGPT Voice in the company’s desktop application. Earlier voice assistants typically waited for a separate turn detector to decide that a user had finished speaking before producing an answer. This could cause unwanted interruptions or noticeable pauses. GPT-Live instead uses a full-duplex architecture that continuously processes incoming audio and can continue listening while it speaks.

The EU AI Office has outlined how monitoring general-purpose AI models after their release could strengthen assessments of systemic risks that may not be fully identified before deployment. The issue was discussed during the fourth meeting of the signatory taskforce for the EU’s voluntary General-Purpose AI Code of Practice, held on 17 July. The meeting covered the code’s Safety and Security and Copyright chapters.

AI was involved in some capacity in 55% of cybercrime cases reported by surveyed African countries in 2025, according to INTERPOL’s latest regional cyber threat assessment. The report draws on responses from 36 of 49 African member countries, alongside operational intelligence, national incident reports and data provided by private cybersecurity companies. Respondents said AI was used occasionally in 47% of cases and frequently in 8% of cases.

Article 16 of the DSA requires hosting services to provide accessible reporting mechanisms. User notices must include a sufficiently substantiated explanation of why the content is considered illegal and identify its precise electronic location. One group reported content without AI assistance. A conventional AI assistant recommended a single legal provision and provided a short explanation.

More than three-quarters of Australian children aged 10 to 17 have used an AI assistant, according to research published by the country’s eSafety Commissioner. The findings are based on an online survey of 1,950 children conducted between 2 February and 4 March 2026. Around 78% had used an AI assistant, while 8% had used an AI companion designed to provide friendship or emotional support. ChatGPT was the most widely used service, reaching 70% of surveyed children.

The White House has finalised a voluntary framework for assessing the advanced hacking capabilities of frontier AI models and invited major developers to discuss how it will operate. Meta, Anthropic, OpenAI and Google were reported to have been invited to staff-level talks with government officials. The White House has not confirmed which companies will attend or whether any have agreed to submit models for testing.

Democracy Reporting International identified the channels while monitoring social media during Hungary’s 2026 elections. Seven had previously been documented, while the researchers connected another 12 to what they call the ‘Common Sense network’. The channels use similar names and almost identical branding, but do not publicly link to one another.

Twenty-eight public and private sector institutions have completed real-value cross-border transactions worth approximately CHF 800,000 using the Project Agorá tokenised payment prototype. Led by the Bank for International Settlements and the Institute of International Finance, the July trial covered 17 transaction scenarios across selected currencies. Individual payments ranged from CHF 9,000 to CHF 125,000 or their equivalents in other currencies.

China’s cyberspace regulator has directed online platforms to take action against more than 1,840 accounts as part of a nationwide campaign targeting misconduct in group entertainment livestreams. Launched in July, the two-month campaign focuses on competitive ‘PK’ broadcasts in which streamers compete against one another while encouraging viewers to send virtual gifts. Regulators are targeting sexualised content, manipulative donation practices and violations involving minors.

The Dutch Data Protection Authority has outlined how the new EU transparency requirements will affect people who use chatbots and encounter content created or altered with AI. From 2 August 2026, organisations must generally make it clear when customers are communicating with an AI assistant rather than a person. The requirement could apply to automated services provided by retailers, healthcare organisations, municipalities and other institutions.