AI Industry Development & Policy Dynamics

Solayer’s founder reveals safety risks of large language models, pointing out that LLM agents relying on third-party API routers face a risk of being attacked by malicious code. Testing shows that multiple routers have security vulnerabilities, and can even leak sensitive credentials. In addition, research demonstrates feasible attack methods and defense measures.

Gate News message, on April 10, according to Gate market data, as of the time of publication, TAO (Bittensor) is trading at $269.80, down 16.28% over the past 24 hours. It reached a high of $341.10 and fell to a low of $248.60. The 24-hour trading volume is $14.71 million. The current market cap is approximately $2.59B. Bittensor is an open-source protocol that supports a decentralized, blockchain-based machine learning network. Machine learning models train collaboratively and earn rewards in TAO based on the value of the information they provide to the collective. TAO also enables external access, allowing users to extract information from the network and adjust network activity according to their own needs. --- Key recent news about TAO: 1️⃣ **

Google has introduced a “Notebooks” feature in the Gemini app, letting users centrally manage chat history and files and sync with NotebookLM to improve workflow efficiency. Currently, it’s only available to subscribers; free users will need to wait. This feature makes it easier for users to organize and use their data, and signals that more updates are coming in the future.

xAI, owned by Musk, filed a lawsuit in Colorado seeking to block SB 24-205 from taking effect. The bill is designed to prevent AI systems from producing algorithmic discrimination. xAI alleges that the law infringes on freedom of speech, requires it to adjust its information output based on specific political stances, and points out logical contradictions within the bill. This is xAI’s second legal action on AI regulatory issues, and it could affect the development of the AI regulatory landscape across the United States.

Bittensor co-founder Jacob Steeves responded to Covenant AI’s exit from the network, proposing to restart the subnet governance mechanism and bring in community votes to elect the team. He believes this is an opportunity to redefine the subnet’s form and criticized Covenant AI’s founder’s decision-making.

Alibaba confirms it has launched its own video generation model, HappyHorse-1.0, developed by the AI innovation division of Taotian Group. HappyHorse-1.0 performed exceptionally well in third-party evaluations. It has not yet been open-sourced and is consistent with Alibaba’s recent closed-source strategy. The release of the model is intended to respond to competitors’ market performance and strengthen its competitiveness in the MaaS market.

Anthropic announced on April 9 that Claude Cowork is officially available for all paid plans, and it added management features for enterprise users, including role-based access controls, group spending limits, and usage analytics dashboards. This product can be used in macOS and Windows desktop applications. Focused on enterprise needs, Anthropic continues to expand its AI product line, from developer-focused tools to team and enterprise-grade application technology.

Andrej Karpathy said that the public’s understanding of AI capabilities has developed a split, divided into two groups: ordinary users using the free version of ChatGPT and professional technical people using advanced models on a paid basis. He believes that the two are discussing entirely different products and emphasizes that progress in the technical field is significant because of verifiable reward functions and incentives driven by commercial value.

Anthropic launched the “Advisor strategy” on April 9, pairing Opus as an advisor with Sonnet or Haiku to run tasks, lowering costs and improving performance through the Messages API. The strategy separates reasoning and execution, allowing executors to consult Opus when they encounter obstacles, achieving higher intelligence in a single API request. Evaluations show that Sonnet paired with Opus improves performance by 2.7%, and reduces task costs by 11.9%. This strategy gives developers a better cost-performance option and has been met with strong enthusiasm from the community.

Gate News update: On April 10, Uniswap founder Hayden Adams posted on a social platform today, asking whether there are any Anthropic (an AI technology company) related resources for coordination, and said he hopes to test Mythos on Uniswap.

Musk’s xAI has recently filed a lawsuit against Colorado’s AI regulations, arguing they violate freedom of speech. The regulation is intended to prevent AI systems from exhibiting algorithmic discrimination across various fields; however, xAI’s own chatbot, Grok, has repeatedly generated racist remarks, highlighting a logical contradiction. This case may become an important precedent in U.S. AI regulation.

Anthropic launched Claude Managed Agents, providing a fully managed AI agent development platform that simplifies enterprise automation workflows and claims it can accelerate product production by 10x. The platform features a secure infrastructure, supports long-duration execution and multi-agent coordination, and helps developers focus more on core business. Notion was among the first to adopt this service, improving work efficiency and showing that competition in the enterprise AI market is intensifying.

Gate News message, April 10, according to multiple people familiar with the matter, in recent internal communications DeepSeek’s founder, Liang Wenfeng, of AI large model company DeepSeek, revealed that DeepSeek’s next-generation flagship large model, DeepSeek V4, will be officially released in late April.

The AI memory system MemPalace, developed with the participation of Milajovа Vіchy, sparked controversy after its test results were questioned for cheating. Real-world testing results show that its efficiency was exaggerated and errors were frequent. The development team has admitted the problem and started fixing it, but multiple data sets still haven’t been corrected, leading to community dissatisfaction.

Covenant AI announced on April 10 that it was withdrawing from the Bittensor network, accusing key figure Jacob Steeves of betraying the promise of decentralization and citing multiple suppression actions. Covenant AI emphasized that the apparent decentralized governance of Bittensor is actually controlled by a single person, creating a “decentralization theater.” The team’s decentralized LLM pretraining project, Covenant-72B, has significant influence in the industry; this withdrawal is seen as a proactive choice rather than a failure.

Research shows that multiple reputable AI benchmark tests have security vulnerabilities that can be systematically exploited to achieve high scores. The research team revealed structural flaws and developed the WEASEL scanning tool to identify and patch these vulnerabilities. They noted that improper evaluation design may distort results and affect the assessment of an AI’s real capabilities.