2026-09-15
AI Briefings Daily Report.

404 Media Exposes OpenAI's Lily Project: Manual Review of ChatGPT Chat Logs to Optimize Models ↗

OpenAI's 'Lily Plan' exposed; this internal project optimizes models by manually reviewing ChatGPT conversations. Reviewers earn over 50 US dollars per hour, but user chat privacy raises broad concerns. Although data is anonymized, leakage risk remains. AI companies' manual review mechanisms are far more common than imagined.

Pony.ai together with GAC releases L4 autonomous driving electric heavy truck T9 Robotruck ↗

Manage Cookie Consent We use technologies like cookies to store and/or access device information. We do this to improve browsing experience and to show personalized ads. Consenting to these technologies will allow us to process data such as browsing behavior or unique IDs on this site. Not consenting or withdrawing consent, may adversely affect certain features and functions.

Trail of Bits criticizes 1Password's AI patch benchmark as misleading, and releases two patch verification Agent skills ↗

1Password's benchmark misrepresents AI patching. We share real-world data on human and agent patch quality from our consulting work and Patch the Planet and release two agent skills for testing and reviewing security fixes.

How many F-Droid apps are LLM-generated? Author reviewed 102 apps and found 72.5% mostly written by AI ↗

Investigating the use of LLMs in the Android FOSS community

Apple launches beta version of Siri AI rebuilt on Google Gemini, temporarily unavailable in the EU ↗

Apple is shipping its rebuilt "Siri AI" after years of delay, built on Google's Gemini models and running partly on the device, partly through Private Cloud Compute. Early testers praise multi-step requests and screen context, but report hallucinations and gaps with personal context. In the EU, the assistant stays unavailable for now.

US Senate proposes new bill requiring AI companies to prove risk prevention measures taken ↗

According to Reuters today (15th), a US Senate aide and a lobbyist familiar with the negotiations revealed that parties negotiating in the US Senate are discussing an AI regulation bill. The bill would require AI companies to prove they have taken reasonable preventive measures to prevent their tools from causing harm.

Musk Speaks: Grok 4.9 May Match Astra / Fable; Grok 5 Could Achieve AGI ↗

Musk outlined the Grok model roadmap, stating that version 4.9 is expected to match top models from OpenAI and Anthropic, while Grok 5 will achieve AGI. This statement has drawn widespread attention.

The Cyberspace Administration of China reports typical law enforcement cases, involving web page tampering, data leakage and violations of AI labeling and security assessment ↗

A relevant official from the Cyberspace Administration of China stated that cybersecurity, data security and personal information protection concern national security, social public interests and the legitimate rights and interests of the people. Relevant units and enterprises must strictly implement legal and regulatory requirements, earnestly fulfill primary responsibilities, and build a strong cybersecurity barrier in cyberspace.

How a children's hospital uses open-source NVIDIA AI for cardiac care ↗

Children’s Hospital of Philadelphia is using open source AI tools to model children’s hearts in seconds — with the goal of enabling safer, more precise care for kids with congenital heart disease.

The National Supercomputing Internet has cumulatively launched over 500 agents, covering over 170 core computing scenarios ↗

The National Supercomputing Internet platform has built six major agent matrices, covering over 170 core computing scenarios. Users can assemble exclusive scientific computing agents like building blocks without programming. The platform's user scale has now exceeded 1 million.

Douyin released"Trustworthy Friendly Community"Plan, launch"AI Truth-Seeking"Fact-Checking Entry Points and Strict Control Over AIGC Misuse ↗

Douyin announced today the launch of the 'Trusted Community Guardian Plan' and the 'Friendly Community Co-building Plan', aiming to build a trusted and friendly community environment through AI+ system governance, optimized self-declaration mechanisms, and strict control of AIGC abuse.

Anthropic and OpenAI Propose Coordinating to Slow Down Frontier AI Development; Critics Like Cohere CEO Question Their True Motives ↗

OpenAI, Anthropic, and Google want to slow down frontier AI development, citing safety concerns. But critics from across the industry and politics are pushing back. Cohere CEO Aidan Gomez calls the initiative a "cartel by another name" designed to shut out competitors, and the White House and Trump himself are opposing the move.

CrowdStrike CEO opposes slowing down frontier AI development, says Pandora's box is already open ↗

CrowdStrike CEO George Kurtz clearly opposed the idea of slowing down frontier AI development proposed by Anthropic CEO in an interview, he believes open-source models are already in circulation, tightening development does not help address threats, and called for strengthening AI runtime security protection.

StepFun releases StepAudio 3 series voice large models, including Realtime, ASR, TTS, Gen, and Music ↗

StepFun officially released the StepAudio3 series of models today, covering five major products including real-time voice interaction, speech recognition, and voice generation. Several ranked first globally on the Artificial Analysis leaderboard. All five models are now live on the StepFun Star Open Platform. This marks a new milestone in AI voice interaction capabilities.

"Big short"Burry on tech giants calling to slow AI development: plainly still profiting for themselves ↗

Investor Michael Burry, known for 'The Big Short', believes these calls are just profiting for themselves. He called the slowdown proposals of leaders of AI companies such as OpenAI and Anthropic 'acts of seeking self-interest'.

WeChat employee debunks rumors of AI assistant Xiaowei directly reading chat records and peeking at privacy: all false ↗

WeChat employee @Kecun Xiaojiang just posted about the WeChat AI assistant Xiaowei and clarified the privacy issues of the feature. He said that some posts claiming Xiaowei "directly reads chat records" or "peeks at all privacy" are false.

StepFun releases StepAudio 3 series voice large models; several rank first globally on Artificial Analysis leaderboard ↗

UK plans to extend permanent residency waiting period from five years to ten years, AI talent begins recalculating future ↗

Checking the site connection security

Former Google DeepMind employee Bilal Chughtai warns AI has the capacity to exterminate humanity ↗

Former Google DeepMind employee speaks out again warning AI may exterminate humanity; previously multiple AI researchers expressed similar concerns, the industry calls for slowing frontier AI development, the statement gained support from Musk and others, Trump disagrees.

Bill Gates: No government is prepared to deal with the social changes brought by AI ↗

Bill Gates, who was originally optimistic about AI, has now shifted his position, warning that AI poses risks such as impacting employment, security, and addiction, saying governments of various countries are underprepared. At the same time his foundation will invest 1 billion US dollars to advance AI for the common good.