An address associated with the Ethereum Foundation has once again sold off 1,000 ETH.

Beating AI Flash News: Cursor has released the findings of its investigation into the earlier poisoning incident affecting the Claude Code plugin. After a two-week probe, the company discovered the attack originated from its AI service provider, Fireworks. According to Cursor, Fireworks leaked an access key granted by the company. Hackers used this key to copy historical Coding Agent operation logs stored by Cursor in shared storage. The data primarily came from 2025 user sessions that did not have privacy mode enabled, originally intended for model training. More seriously, some API keys in these records were not redacted. Using these credentials, hackers further breached some internal software services used by Cursor, accessing archived non-production code, customer support emails, product usage data, partial AI conversation summaries, and credentials users had submitted to support. On October 29, hackers used credentials stolen from Cursor internally to tamper directly with the extension store’s cache, distributing versions Claude Code 2.1.312 and 2.1.313 embedded with an SSH backdoor to users. The malicious versions were available for approximately 11 hours, with up to 5,000 users potentially downloading them. The upstream plugin marketplace Open VSX was not tampered with, and Anthropic’s official versions remained unaffected. Cursor disclosed that the leaked Coding Agent logs involved fewer than 0.05% of its users, with fewer than 0.01% of users having their API credentials exposed. Sessions with privacy mode enabled, customer code repositories, and production systems were not impacted. Cursor noted the last detection of hacker activity was on November 1, when it was successfully intercepted. The company has now notified affected users, fully ceased using Fireworks for inference requests, revoked its data access permissions, rotated internal credentials, and redeployed production infrastructure.

According to EmberCN's monitoring, the address that went long 1.38 million HYPE at $38.6 last November and has held the position ever since actually belongs to @smartestmoney. The position generated an unrealized profit of roughly $82 million when HYPE rose to $98 in September this year, but as the token has recently dipped below $70, that unrealized gain has shrunk to about $42.95 million, nearly halving. Given the address had previously withdrawn a large amount of margin, the current liquidation price of the position stands at $65.15, just $4.65 away from the current market price. If HYPE continues to decline, this long position will face liquidation risk.

Insight Beating AI News: Anthropic has updated Claude Code Projects, with a key focus on boosting the initiative of its orchestrator agent. The agent can now answer questions directly in the main project chat, proactively report progress on various tasks, eliminating the need for users to open individual sessions to check updates. Two existing commands have also been added to Projects: `/btw` lets users move emerging new issues to a separate session without interrupting the current task; `/fork` creates a new session from an existing one to test different solutions, leaving the original work untouched. Users can also set reminders for any session—for example, asking Claude to remind you to check a specific task tomorrow will temporarily hide the relevant session from the list, which will reappear at the scheduled time and be flagged by the orchestrator in the main chat. Additionally, individual sessions can now generate read-only links to share with any logged-in Claude user, though sessions accessed from a local computer cannot be shared. The interface has also added entry points for recently used, updated, and pinned sessions, plus thumbnails of files generated in each session; project settings and local folder access have been simplified. These updates are available on web and desktop, with the latest mobile version adopting the new interface. Claude Code Projects remains in public beta for Pro and Max users; users who have not yet qualified can continue to apply for the waitlist.

Crypto analyst Murphy noted in a post that following yesterday’s sharp BTC plunge, the high-trapped coins of short-term holders (STH) only saw minor loosening. Data shows roughly 1.48 million BTC are held in the $73,000–$86,000 range, making up 68% of total STH holdings, all currently in unrealized losses. Yet only a net 34,000 BTC were removed from this range yesterday, accounting for ~2.3%, indicating most short-term holders did not panic-sell amid the downturn. Specifically, the $75k, $80k, and $85k cost ranges saw reductions of 7,282, 8,481, and 5,441 BTC respectively, with selling pressure relatively dispersed. Meanwhile, total STH holdings added a net 17,500 BTC that day, meaning some long-term holders (LTH) also distributed their coins during the drop. Coin accumulation was concentrated mainly at $68k, with ~42,200 new BTC added to this range alone. For profitable coins, the $62k–$64k range holds ~480,000 BTC, the only large profitable segment among current STHs, with a reduction of just ~3,100 BTC yesterday, less than 1%. Murphy argues this range is the nearest immediate support below; if BTC falls into it, short-term holders’ profitable coins will largely be erased, and market sentiment pressure could rise further. Conversely, if the upper trapped coins stay stable, prices still have a chance to regain support going forward.

Beating AI Flash News: Google is launching a dynamic workflow feature called `/run-workflow` for Antigravity. Users only need to describe a task, and the AI will automatically generate a Python script to orchestrate multiple agents to execute it. They can predefine which tasks run in parallel and which must be completed sequentially. For example, if a user asks the AI to audit a large codebase, it can assign multiple agents to review different modules simultaneously, pass identified issues to a fix agent, and finally run unified tests. Each agent’s output is passed to subsequent steps in a specified format, reducing information loss and execution misalignment during task runs. The generated script can be reviewed and modified by users before execution, or saved for reuse. For similar tasks later, users can directly run the same workflow without redefining the division of labor each time. Currently, `/run-workflow` is being gradually rolled out to Google AI Pro and Ultra users of the Antigravity App and CLI.

Beating AI Express News: Anthropic announced that Claude Duet has extended invitations to all US Pro, Max, and Ultra subscribers who were previously on its waitlist. New users may apply for access, but the feature is not yet available outside the United States. To encourage trial usage, Anthropic is offering additional free quota resets: 3 for Pro users, and 1 each for Max and Ultra subscribers. Each reset restores a user’s full weekly Claude usage quota and resets the applicable 5-hour session limit. The reset can be manually activated via Settings > Usage in Claude’s web or desktop versions; unused resets will expire on December 15. Claude Duet carries no separate fee, and shares its subscription quota with Claude Chat and Claude Code. Even when no active tasks are issued, background operations such as checking emails and calendars will consume quota. Anthropic stated it plans to roll out Claude Duet to more countries outside the US soon.