SlowMist’s Yu Xian: Stolen funds from SecondFi may exceed $20 million.

According to PeckShield monitoring, the wallet address identified as the attacker of Aztec Network’s Private Rollup Bridge has once again deposited 300 ETH into Tornado Cash, valued at approximately $572,000. To date, the attacker has transferred a total of 500 ETH to Tornado Cash. Earlier, Aztec Network suffered a security breach in June 2026, resulting in losses of around $2.165 million in crypto assets. This fund transfer may further complicate the tracing of stolen assets.

According to HTX market data, TUT has rallied more than 55% in the past 24 hours, currently trading at $0.04838, with its market capitalization rising to $40 million.

International Monetary Fund (IMF) First Deputy Managing Director Dan Katz said local stablecoins designed to reduce reliance on USD stablecoins may actually accelerate users’ shift toward USD stablecoins. Katz noted that when local stablecoins and USD stablecoins operate on the same blockchain infrastructure, users can swap between them via decentralized exchanges, liquidity pools, or peer-to-peer transactions, which could lower capital conversion costs and shift foreign exchange activities from traditional banks and currency dealers to on-chain platforms. “Local stablecoins could even accelerate the adoption of foreign exchange stablecoins (USD stablecoins),” he said. Citing South Africa as an example, Katz pointed out that while USD stablecoins already have some local adoption there, demand for local stablecoins pegged to the rand is lower. While no definitive conclusions can be drawn yet, users may prefer USD stablecoins due to their higher liquidity, stronger network effects, and broader acceptance across platforms and borders. Katz argued that the impact of stablecoins varies by country context: in highly dollarized economies, stablecoins may primarily replace existing USD-denominated assets; in countries with limited access to USD and weaker economic fundamentals, stablecoins could further increase demand for foreign currency. He called on global regulators to include stablecoin on-ramps, off-ramps, and on-chain trading platforms in their regulatory frameworks to mitigate potential risks.

U.S. sources told media on the 7th that U.S. Chairman of the Joint Chiefs of Staff Dan Caine is seeking ways to disengage from the Iran conflict. Three people familiar with the matter disclosed that over recent weeks, Caine has privately made clear to other senior Trump administration advisors that the U.S. needs a path out of the Iran conflict, as military options on the table could backfire, and air power alone is unlikely to fulfill Trump’s goals. Sources added that Caine has discussed military options for escalating the conflict with other government officials—including Vice President Vance, Secretary of State Rubio, and CIA Director Ratcliffe—while also proposing avenues to end the hostilities. Recently, Caine has also held private meetings with several like-minded Trump advisors to ensure they reach a consensus ahead of their meetings with the president. Additionally, during his recent talks with Trump, Caine voiced concern over the U.S.’s dwindling ammunition reserves, and the two sides also discussed potential options to escalate the conflict.

A developer recently open-sourced the kimi-k3-in-c project, attempting to run the 2.78-trillion-parameter Kimi K3 model on devices with only 8GB of RAM. The project is just 176KB in size, written purely in C99, and does not depend on GPU, CUDA, PyTorch, or BLAS—its model inference can be completed entirely via CPU. This solution leverages the Mixture of Experts (MoE) architecture features of Kimi K3. While the model’s total parameter scale reaches 2.78T, only 16 of the 896 experts per layer are activated. Instead of loading the full ~1.56TB model weights into memory, the developer stores most expert weights on NVMe drives and reads them in real time as needed for inference; some dense trunk layers also adopt layer-by-layer streaming loading. However, the solution still has obvious performance limitations: in 8GB RAM mode, the model takes approximately 32.7 seconds to generate one token, and requires nearly 1.7TB of high-speed storage support. The developer stated that this solution is currently more of an experimental exploration into optimization directions for large model inference infrastructure, and does not have practical production use value. Nevertheless, its approach of "NVMe streaming loading + MoE sparse activation" provides a new idea for the future low-cost operation of ultra-large models.

JPMorgan said the Q2 earnings season is drawing to a close, with roughly 80% to 85% of companies in the U.S. and Europe having reported results so far, and overall earnings performance has come in significantly stronger than expected. Data shows that both the U.S. and Europe have seen widespread EPS (earnings per share) beats, with the proportion of companies exceeding market expectations rising notably from prior periods. Despite already high market valuations and earnings expectations ahead of the earnings season, the combined EPS of S&P 500 constituents has continued to be revised upward, while the share of companies cutting their profit outlooks has fallen to its lowest level since 2021. By region, corporate earnings growth in the U.S. and Europe hit 25% and 23% year-over-year, respectively, both beating consensus market expectations; revenue growth rose 14% and 10% YoY, respectively, indicating robust top-line growth for companies. By sector, JPMorgan noted that the energy sector is one of the key drivers of earnings growth in both regions, fueled primarily by rising energy prices due to geopolitical conflicts. In addition, the financial and technology sectors also made significant contributions to overall profit growth.