Anthropic whistleblower takes AI crisis to Capitol Hill, while Trump and Jensen Huang call it a hoax from the other side. Crypto bill falls one vote short, two years of legislation down the drain.
1 | AI safety battle spills from inside companies to Capitol Hill, whistleblower says company is "betting all our lives"
Former Anthropic researcher Jacob Coxon publicly posted after resigning, saying frontier AI companies are "betting all our lives," and the article quickly spread on social platforms. Anna Wang of Anthropic's AGI safety team subsequently said many people inside the company want to slow the pace of development. This is the first time the safety debate at a frontier lab has publicly spilled out from insiders.
Meanwhile, executives from OpenAI, Anthropic, and Google have held weeks of closed-door negotiations to discuss limiting development speed and establishing a new safety oversight body. Amodei published a long essay on "slowing the frontier," Altman expressed agreement, and OpenAI's IPO was delayed to 2027. The three leading companies have rarely converged on safety positions, forming a diagonal with the White House. Trump dismissed AI risk as a "hoax," and Jensen Huang called safety panic "pure fabrication."
Yesterday was a battle of statements between CEOs; today's turning point is the whistleblower. When internal employees begin testifying to Congress, the debate changes—it is no longer "whether to slow down," but "whether frontier labs' safety commitments are credible."
(Sources: NPR / TechCrunch / Fortune / NBC News)
2 | Sanders and Bannon sit at the same table, demand permanent ban on superintelligence
Bernie Sanders and Steve Bannon appeared simultaneously at a "Pro-Human Assembly" event on Tuesday. Sanders announced he will jointly propose legislation with Representative Casar to permanently ban the development of superintelligent AI. He also proposed a 32-hour workweek bill tied to AI and said he hopes Trump and Xi Jinping can reach an agreement on an AI pause at the upcoming summit.
Bannon approached from a nationalist angle, calling AI the "decisive issue," but what concerns him is not human safety but technological sovereignty. The two have completely different answers to "why regulate": Sanders worries about workers being replaced, Bannon worries about the nation being hollowed out.
The far left and far right of the political spectrum are simultaneously opposing unrestricted AI development, and the center has lost its exclusive say over tech policy. But "opposing the same thing" does not equal "agreeing on what to do." Sanders wants an international treaty; Bannon wants America First. The coalition's symbolic significance is as noteworthy as its fragility.
(Sources: Axios / Fortune / NPR)
3 | Clarity Act Fails 49-50, Crypto Market Structure Legislation Goes to Zero
The Senate procedural vote results are in: 49 in favor, 50 against, 11 votes short of the 60-vote threshold. Four Republicans crossed the aisle to vote no, including Collins, Hawley, Moran, and Tillis. Elizabeth Warren expressed support for crypto regulation, but "not this version." After the vote, crypto stocks fell across the board, and Cathie Wood reduced crypto-related holdings that day.
Trump made over 120 ethics concessions, including setting a timeline for his $800 million token holdings in World Liberty Financial, and it still wasn't enough. The bill took a year to go from House passage to Senate stall, with the industry investing years of time and hundreds of millions of dollars in lobbying expenses.
The Clarity Act was supposed to clarify, for the first time at the federal level, the SEC and CFTC jurisdictional divide over crypto assets. Its failure effectively ends crypto market structure legislation for 2026. The lesson is straightforward: crypto regulation cannot advance as a standalone issue and must hitch a ride on a larger legislative package. But with the AI safety debate consuming Congress's attention, the opportunities for hitching a ride are also shrinking.
(Sources: CoinDesk / Al Jazeera / Fortune / NPR)
4 | China-U.S. Summit AI Agenda Emerges, Chips and Slowing Down May Be Bundled Into Talks
The Chinese and U.S. heads of state will meet formally on the 24th. The agenda list is already semi-public: Nvidia chip export restrictions, AI competition rules, EV tariffs, rare earth supply. One possible concrete outcome is joint monitoring of AI-driven cyberattacks by both countries. J.P. Morgan estimates 2026 hyperscale data center capital expenditure at $697 billion, and the direction of the chip supply chain directly affects how this investment is distributed between China and the U.S.
AI safety is no longer just a debate between Silicon Valley and Capitol Hill—it has become an item on the U.S.-China bilateral negotiating table, discussed alongside trade, tariffs, and supply chains. Chip export controls and calls for an AI slowdown could be packaged into the same deal.
(Sources: Fortune / SCMP / Rest of World / J.P. Morgan / Dallas Fed)
5 | ByteDance's profits decline, all non-AI businesses switched to "bookkeeping mode"
ByteDance's profit fell to approximately $20 billion in the first half of 2026, a slight year-over-year decline. Meanwhile, 2026 AI capital expenditure was raised by 25% to over $29 billion.
The more critical signal is the organizational restructuring: ByteDance has reclassified all non-AI businesses into "bookkeeping mode," requiring these businesses to achieve positive cash flow. This is not "All in AI" at the slogan level—it is splitting the company in two at the management level. The AI side invests regardless of cost, while the rest is only required to survive and generate cash. TikTok, Douyin, and Toutiao—once core product lines—have become blood-supply machines for AI investment under the new framework.
ByteDance is the first tech giant to declare AI priority through organizational structure. Google, Microsoft, and Meta are all increasing AI spending, but their non-AI businesses are still discussed by management as growth engines. ByteDance's approach is more extreme—growth of non-AI businesses is no longer on management's agenda.
(Source: Bloomberg)
Also worth knowing ↓
Salesforce and Nvidia jointly release reasoning model Koa. The two major companies trained and released a reasoning model together for the first time, as reasoning capabilities are spreading from the exclusive domain of OpenAI and Anthropic to a broader alliance. (Source: TechCrunch)
Gates Foundation pledges $1 billion to combat AI inequality. The funds will be used to help low-income countries access AI tools. Gates also compared AI to "alien intelligence," warning that governments are not prepared. In a week dense with calls for a slowdown, this is one of the few voices focused on "what to do with AI" rather than "whether to stop." (Source: Fortune)
Carlyle executive warns AI financing model resembles pre-subprime crisis mortgages. Dallas Fed data shows banks' AI exposure accounts for only 0.8% of assets, with GPU-backed loan rates at approximately 11%. The systemic risk may not be on the banking side, but rather with small and mid-sized AI companies that use GPUs as collateral and rely on inference revenue to repay loans.
AIUC (founded by former Anthropic employees) raises $40 million, focusing on constraining runaway AI agents. The founding team came out of Anthropic, choosing to build a "brake" rather than a model. On the same day as the whistleblower incident, former employees expressed the same judgment through capital markets. (Source: TechCrunch)
Elon Musk is living in an Airstream trailer to oversee the xAI Colossus expansion. A classic scene of Musk-style management: the CEO lives on the construction site. The Colossus that xAI is building is currently one of the largest single AI training clusters. (Source: Fortune)
DOJ charges former Robinhood engineer with front-running on Hyperliquid. The U.S. Department of Justice has filed its first criminal charges for front-running on a decentralized exchange, a signal that carries more significance than the case itself. (Source: CoinDesk)
Liang Wenfeng's CFO is in place, previously invested in Zhipu and MiniMax. DeepSeek's personnel puzzle is being completed. A CFO with an AI investment background means financing and an IPO may already be in the plans. (Source: QbitAI)
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
