OpenAI Discovers Critical Brotli Vulnerability; Bitcoin ETF Experiences Significant Outflow

ME AI message, 1⃣ OpenAI's red team discovered an unauthenticated remote code execution vulnerability in the Brotli compression library; Node.js released a patched version on October 21, and bundled software such as Bun and Envoy also require updated versions.
2⃣ The UK has committed £100 million to build an AI biosecurity lab, with support from OpenAI, Anthropic, and Google DeepMind, to enhance the government’s ability to understand and respond to AI-related risks.
3⃣ Bitcoin spot ETF funds experience significant reversal; Farside data shows $635.6 million net outflow from U.S. spot BTC ETFs on October 22, the largest single-day outflow since June 25.
4⃣ The Initia chain is suspected of suffering an "infinite minting" attack; the attacker has cashed out approximately $2.6 million, and chain addresses still hold around $4.54 million in assets. The official team has not yet responded.
5⃣ The final U.S. Consumer Confidence Index fell to a five-month low of 48, with one-year inflation expectations rising to 4.7% and five- to ten-year inflation expectations rising to 3.5%.
6⃣ Attacks on Middle Eastern vessels and risks to energy infrastructure have increased the oil price risk premium; Goldman Sachs estimates the Brent risk premium at approximately $27, with a base forecast of $90 by year-end and an average of $80 in 2027.
7⃣ Franklin Templeton completes the final regulatory steps for two Bitcoin dividend ETFs; products such as FBEQ will initially allocate 95% to stocks and 5% to Bitcoin, using stock dividends to purchase Bitcoin.
8⃣ Sentora’s review shows that JPM Coin on Ethereum has zero supply and zero transfer records; the contract has only one allowlisted address and holds no tokens, indicating that actual activity has not yet begun.
9⃣ OpenAI stated that an internal research model briefly gained external internet access due to a DNS configuration error, leading to a 11-day pause in frontier training, and has since incorporated high-compute red teaming into its training safety evaluations.
10. Research by Meta and Carnegie Mellon University shows that an 8-billion-parameter model trained with reinforcement learning outperformed the larger Llama-4-Maverick in tool selection tasks, highlighting a shift in AI development from scale to efficiency.
1⃣1⃣ According to Japanese central bank sources, further rapid rate hikes will continue, with particular focus on December; current rates still have a significant gap from neutral rates, and inflation needs to stabilize near the target.
1⃣2⃣ Block releases the full-duplex voice dialogue model Dessa research preview, trained solely on human speech, capable of speaking and listening simultaneously, and running locally in real time on devices such as the MacBook Pro M4 Max. (Source: Followin Flash News)