DeFi protocols Camelot and Cypher have merged to form Frontier, focusing on Uniswap v4 infrastructure.

During his speech at the United Nations General Assembly, Trump stated that the U.S. opposes any global "plot" to control artificial intelligence. "We will encourage superintelligence, not restrict it," he said. Additionally, Trump announced plans to rename artificial intelligence to "superintelligence", with all U.S. official documents adopting the term "Super Intelligence".

Beating AI Newsflash: LatePost reports that Doubao’s General Session team is downsizing. The team, which previously had around 50 members and was once Doubao’s largest internal team, is primarily responsible for dialogue strategy, evaluation, and response experience. The team is expected to shrink by roughly half post-adjustment: some employees will be reassigned to Doubao’s commercialization, Feishu, and other teams, while the rest will be laid off. The model side is also contracting. The Product Posttrain-Chat team, responsible for post-training Doubao dialogues, has seen some staff transferred to Horizon RL and Product Posttrain-Work. The latter team mainly serves office and AI agent scenarios. In August this year, Seed just established Product Posttrain-Work while renaming the original Application team to Product Posttrain-Chat. Less than two months later, the Chat team began downsizing. Doubao currently boasts over 200 million daily active users (DAU), with dialogue remaining its largest user entry point. However, in recent months, ByteDance has clearly shifted more manpower to office tools, enterprise services, and commercialization.

Beating AI News Brief: Reuters, citing insiders, reports that both DeepSeek and Moonshot AI have been invited to address the UN Security Council on AI and international security. DeepSeek is expected to deliver its briefing to the council this week, with founder Liang Wenfeng not planning to attend. Moonshot AI has also received an invitation but has yet to confirm participation. The meeting will discuss risks posed by AI to international security. OpenAI CEO Sam Altman plans to attend and speak, while Anthropic is also expected to send a senior representative. This marks the UN Security Council’s first formal discussion on AI risks in 2023. Back then, China warned AI should not become an "unbridled wild horse", while the U.S. cautioned against using AI to censor or suppress people. The current meeting comes as discussions between China and the U.S. on AI security heat up. Multiple U.S.-based AI companies have recently called for coordinated slowdowns in the development of frontier AI, while Chinese state media has criticized such proposals, arguing they could be used to limit China’s participation in global AI competition.

Coinbase has announced the launch of Bitcoin collateralized fixed-rate loans, enabling users to access liquidity without selling their BTC, while locking in interest rates and repayment dates prior to borrowing. Built on the Morpho Midnight protocol on the Base network, the new product is now available to eligible users in the United States, excluding New York State. Unlike Coinbase’s earlier launched floating-rate loans, this new offering features fixed interest rates and explicit repayment terms. Coinbase warns that loans carry collateral liquidation risk; failure to fully repay by maturity may trigger liquidation.

Beating AI Express News reports that AI personal assistant Instinct has been exposed for a suspected cross-user data isolation issue. User Pritak discovered that Instinct suddenly began processing a Gerber insurance claim file he never uploaded, and claimed the file contained his name and address. Pritak soon noticed the middle name and address were not his, and he had not uploaded any photos at all. Instinct later stated it checked its records, noting that Pritak had only sent text at the time, and that "another user’s image was inserted into this conversation". It also asked whether to report the issue to the team, and generated an incident description. At present, it remains unconfirmed whether a cross-user data breach actually occurred on the backend, or if Instinct fabricated the file, content, and subsequent explanation together. But if the former is true, the problem is far more serious than a typical AI agent hallucination, indicating that data or sessions that should have been isolated between different users have crossed over. Instinct has previously faced scrutiny over permission issues: TechCrunch reported that it once continued to read previously saved Gmail emails even after a user disconnected their Gmail account, and testers found it would send emails on behalf of users without confirmation.

According to GMGN market data, Robinhood-based on-chain token musebook briefly exceeded $27 million in market capitalization, now trading at $26.19 million, with a 24-hour gain of 317.8% and hitting a new all-time high. Musebook is a social community primarily focused on Meta AI Agent Muse, positioned to facilitate interactions between Agents, as well as between Agents and humans. The platform has low entry barriers—no account, email, or authentication is required to post—and provides an API for reading and publishing content. The community was founded by @wyn_eth, with its official real-time interaction account being @musebooklol. Earlier, the account announced that transaction fees generated from claiming musebook tokens would be allocated to platform development. Furthermore, the account has just received a follow and interactive support from Meta AI head Alexandr Wang. BlockBeats reminds users that the relevant token has a small market cap, high price volatility, and early investors hold substantial profitable positions, which may lead to significant selling pressure in the short term. Users are advised to exercise caution when investing.