Jelitics cuts losses by selling half of its SI holdings, with total losses amounting to $562,000.

JPMorgan strategist Nicholas Panigirtzoglou notes that the October sell-off in AI stocks is unlikely to replicate the sharp rebound seen in July. Compared to earlier periods, current market willingness to buy the dip has weakened, short positions are insufficient to fuel potential short-covering buying, corporate buyback funds have shrunk, and the renewed buildup of leveraged exposure in ETFs could further amplify market volatility. JPMorgan believes that based on investor positioning and fund flows, AI infrastructure stocks do not appear to have entered oversold territory yet. The current sell-off has spread from South Korean memory chip stocks to the U.S. semiconductor sector. SK Hynix and Samsung Electronics have both seen their share prices fall by nearly a third over the past month, while the Philadelphia Semiconductor Index has dropped 10% in the last five trading days. Additionally, the combined $40 billion share buyback plan previously announced by SK Hynix and Samsung Electronics was completed ahead of schedule in mid-October, potentially removing some of the buying support from corporate repurchases. Meanwhile, BlackRock CEO Larry Fink warned that a pause in U.S. data center construction could push the economy into recession, noting that a significant portion of U.S. economic growth is tied to capital expenditures on AI infrastructure. As market concerns grow over AI investment returns and corporate profitability, price adjustments in the AI industry chain could also impact data center construction, energy infrastructure, and related employment sectors.

Security firm Hacken reports that in Q3 2026, the crypto industry recorded 107 hacking attacks, with total losses reaching $1.18 billion—the highest quarterly loss since Q1 2025. Of this, roughly $906 million (77% of total losses) was linked to breaches of third-party vendors or development tools, components with publicly disclosed fixes not yet deployed, and un-audited elements. The quarter’s losses were driven by four major incidents: Bitget’s hack caused ~$387 million in losses, Liquid Network lost ~$320 million, the Tectonic incident involved ~$120 million, and a Coldcard firmware bug resulted in ~$115 million in damages. While third-party infrastructure issues accounted for the largest loss sum, smart contracts remained the most common attack vector, with 71 incidents leading to ~$228 million in losses. On loss recovery: Approximately $387 million was retrieved or reversed via rollbacks, representing 33% of total losses; another ~$410 million was covered by operators’ own funds, including the $387 million from Bitget’s user protection fund. Around $384 million (32% of total losses) remained unrecovered and uninsured. Hacken advises the industry to enhance security audits of critical components, including third-party vendors, development tools, oracles, cross-chain bridges, custodial systems, and signing permissions.

According to Cryptoquant data, the cumulative inflow of XRP whales into Binance over the past 30 days has climbed to 1.669 billion tokens, marking a 7-month high and surpassing the roughly 1.65 billion figure recorded in September. This signals a renewed uptick in large holders’ activity of transferring XRP to crypto trading platforms. Increased inflows from whales could boost XRP’s tradable supply on these platforms; if large holders opt to sell, it may intensify market selling pressure. That said, capital inflow does not automatically point to an imminent sell-off—it may also stem from wallet rebalancing or utilization of other Binance trading services. This metric alone is insufficient to determine XRP’s future price trajectory; further analysis combining net flow, trading volume, and price movements is required.

According to HTX market data, Bitcoin briefly rallied to nearly $84,000 on Wednesday before a sustained pullback, falling to around $80,400 by early Thursday, marking a 2.6% 24-hour drop and approaching last Thursday’s low of roughly $80,430. The U.S. September core inflation rate rose to 2.6%, exceeding the market expectation of 2.5%, further stoking Federal Reserve interest rate hike bets. Meanwhile, selling pressure on AI stocks intensified, with Nasdaq 100 Index futures down 1.2%. Data disclosed by Anthropic to potential investors shows its third-quarter revenue grew by $16.8 billion, a figure reflecting a slowdown in growth momentum. The crypto market weakened in tandem: ZEC dropped 11% to around $1,100, while ETH fell 5% to roughly $2,370. U.S. spot Bitcoin ETFs recorded a net outflow of approximately $261 million on Wednesday. Analysts noted that if Bitcoin closes below $80,400 on a daily basis, it will lose the key level that buyers have defended twice since last Thursday.

Beating AI News Flash: OpenAI has included "honey traps" in its internal security training to prevent employees from leaking trade secrets. The Wall Street Journal reported that the company once shared a spy story with staff: a foreign intelligence agency tried to seduce a U.S. diplomat stationed in Russia. It first sent a blonde woman with champagne to knock on the diplomat’s hotel door, which was rejected; then switched to a red-haired woman, and finally even dispatched a young man. OpenAI used this case to remind employees that spies may exploit romantic and intimate relationships to approach targets and gather confidential information. The company also teaches staff to watch out for casual small talk in daily life. Another training example involves a stranger who often appeared at a San Francisco playground with a child, gradually built a rapport with an OpenAI employee, and then probed for work-related details during chats. The company even requires employees to avoid wearing clothes with the OpenAI logo in public, apply privacy screens to their laptops on the subway, and wipe whiteboards clean after meetings to prevent cleaning staff from accessing sensitive information. These trainings were rolled out long before the wiretapping case was exposed this week. On October 13, the U.S. Department of Justice disclosed that an OpenAI engineer was arrested by the FBI on suspicion of installing recording devices in an executive meeting room. Anthropic and Google also conduct similar anti-spy training.

Beating AI Express News: Anthropic has launched the public beta of OpenTelemetry tracing for its Claude Platform. Agents can now chain model calls, data searches, and tool executions, allowing developers to track runtime logs to view each call’s duration, error points, and token usage. This eliminates the need to piece together execution flows from disparate service logs, making it easier to identify bottlenecks that slow down tasks or drive up costs. The most notable update is native API support. While Claude Code and Cowork already had similar monitoring capabilities, the Messages API and cloud-hosted Agents can now directly send runtime logs to platforms including Grafana, Datadog, and Langfuse. Administrators only need to configure the receiving endpoint in the Claude Console, without modifying individual existing API calls. However, tools run by developers themselves still require supplementary tracing records via SDKs or other methods. By default, the system only exports metadata such as duration, status, and token usage. Enabling the Full content feature also logs prompts, model responses, and tool inputs/outputs for error review, but this data is sent to external monitoring platforms, requiring enterprises to separately manage access permissions for sensitive data.