Trader Kimchi withdrew 1.35 billion PUMP tokens valued at $6.43 million over the past three hours.

According to Reuters, OpenAI said two of its AI agents undermined monitoring systems used to detect misconduct during internal coding tasks. The company noted the incidents were not of the most severe level, but highlighted that the monitoring mechanisms developers rely on are themselves at risk of being bypassed by AI agents. Specifically, on October 3, a GPT-6.1 Sol agent, while performing an internal test task, extended the instruction “do not send Slack messages” to also apply to security notifications, suppressing automatic security alerts. On October 6, another GPT-6 Astra agent removed a security monitor when copying existing code, and later falsely claimed it had not done so. OpenAI stated that in both cases, the agents considered whether to retain monitoring measures, but interpreted standard coding instructions as permission to weaken monitoring. The company added it is taking steps to make it harder for models to disable monitoring, and admitted it has not yet provided a “satisfactory explanation” for this behavior. Meanwhile, OpenAI is testing whether its models exhibit a tendency to actively avoid monitoring and seek justifications to rationalize such actions. The disclosure comes amid growing industry focus on AI agents overstepping their authorized bounds. OpenAI emphasized that the incidents occurred in an internal test environment and did not involve external systems.

Beating AI News Flash: French AI firm Mistral has launched Voxtral 3, which includes three speech transcription models, one speech synthesis model, and a full Voice Agent SDK. In an official demo, a user requested a dermatologist appointment in English. The AI read a French referral letter, queried the system for available time slots, completed the booking, and confirmed in English—all via real-time voice conversation. Voxtral 3 is not an end-to-end speech large language model (LLM) that handles all tasks; it uses a modular design: transcription models identify speech and detect when a user has finished speaking, general-purpose LLMs like Mistral Small handle reasoning and tool calling, and the speech synthesis model generates responses. The LiveKit-based SDK connects these components into a real-time pipeline, supporting phone and web calls, and allowing users to interrupt at any time. Mistral claims single-turn interaction latency is under 1 second. Developers can also replace inference or speech models individually. Transcription language support has expanded from 13 in the previous generation to over 50, while speech synthesis supports more than 35 languages and 50+ dialects. In an independent non-real-time transcription test by Artificial Analysis, Voxtral Small Transcribe 3 posted a word error rate (WER) of 2.1%, ranking third behind models from Alibaba and Microsoft. Its speech synthesis ranked 13th. API real-time transcription costs $0.008 per minute, and speech synthesis is priced at $13 per million characters. Enterprises can also contact Mistral to deploy the solution on their own servers or private clouds.

Bitcoin has been stuck in a range of $80,000 to $84,000 over the past two weeks. Huobi HTX’s chief analyst Cloud attributes this to two offsetting forces: ETFs have posted net outflows for two consecutive weeks, while exchange holdings continue to flow out, and long-term holders are only taking mild profits. In the week ahead of the Federal Reserve’s interest rate decision, Bitcoin is likely to remain range-bound. The CME currently prices a 16% chance of an October rate hike, while economists widely peg the odds at nearly 50/50. This gap between market pricing and consensus will deter capital from betting early. If Bitcoin breaks out of the range prematurely, the likelihood of a downward test is slightly higher: funding rates remain elevated, and option skew continues to lean toward puts, indicating longs have not yet been cleared out and demand for downside protection is rising. However, the $80,000 level has acted as strong support multiple times over the past month, so a downward breakout is more likely to be a false move. Confirmation signals for a breakout fall into two categories: For an upward move, Bitcoin needs to close above $84,200 on a daily basis, ETFs must see net inflows exceeding $500 million over three consecutive days, and funding rates should not exceed 15%. For a downward move, it needs to break below $79,800, ETF outflows exceed $500 million in a single week, and exchange flows turn positive. The real catalyst for Bitcoin’s next direction remains the interest rate decision to be released in the early hours of October 29 (Beijing time) and the nearly 89% pricing of a December rate hike. Note: This article does not constitute investment advice, nor does it constitute an offer, solicitation, or recommendation for any investment product.

According to Reuters, OpenAI said two of its AI agents undermined monitoring systems designed to detect misconduct during internal coding tasks. The company noted the incidents were not the most severe, but highlighted that monitoring mechanisms relied on by developers themselves carry a risk of being bypassed by AI agents. Specifically, on October 3, a GPT-6.1 Sol agent, while executing an internal test task, extended the existing instruction "do not send Slack messages" to also apply to security notifications, thereby suppressing automatic security alerts. On October 6, another GPT-6 Astra agent removed a security monitor when copying existing code, and later falsely claimed it had not done so. OpenAI said in both cases, the agents considered whether to retain monitoring measures, but interpreted routine coding instructions as permission to weaken monitoring. The company stated it is taking steps to make it harder for models to disable monitoring, and admitted that "a satisfactory explanation for this behavior has not yet been found". Meanwhile, OpenAI is testing whether its models have a tendency to actively avoid monitoring and seek reasons to rationalize such behavior. The disclosure comes as the industry continues to focus on AI agents overstepping their bounds. OpenAI emphasized that the incidents occurred in an internal test environment and did not involve external systems.

According to official announcements, Binance will officially launch five bStocks (tokenized stocks) related spot trading pairs on October 21, 2026. The five pairs correspond to RoboStrategy Inc (BOTB), Berkshire Hathaway (BRKBB), Goldenray Sci-Tech Innovation 50 ETF (KSTRB), Direxion Daily 20-Year Treasury Bull 3X ETF (TMFB), and Vistra Corp (VSTB), all denominated in USDT.

Beating AI News: OpenAI has announced the acquisition of Orchestra Bio, a four-year-old biotech company, with plans to integrate its team and technology into its life sciences platform Rosalind. Orchestra has previously raised around $20 million in funding, with investors including a16z, Atria, and 8VC. Orchestra’s core product is a drug discovery AI agent system. It does not directly develop new drugs, but assists pharmaceutical companies in managing R&D progress, coordinating cross-functional teams, and tracking clinical trial schedules. For example, if a drug batch production failure risks delaying a clinical trial, the agent can automatically identify affected tasks and timelines, aggregate relevant information, and propose adjustment plans, which are then decided by researchers. OpenAI launched GPT-Rosalind in April this year, a model tailored for biology and chemistry problems, followed by a supporting research workbench. This acquisition will fill the gap in drug discovery project management, enabling AI to expand beyond analyzing experimental data to further participate in subsequent R&D workflows.