Hark, from the founder of Figure AI, is now officially live, accessible on Web, iOS, and Android.

Santiment released data this morning showing that on Monday, Bitcoin’s net outflow from centralized exchanges (CEXs) reached 24,073 BTC, marking the largest single-day outflow since March 1. CEX Bitcoin supply has fallen to approximately 6.50% of the total BTC supply. Santiment notes that large-scale outflows are typically a bullish signal, as assets leaving CEXs become less readily available for immediate selling. If demand remains strong while circulating supply shrinks, buyers will have fewer coins to absorb, which will push prices higher.

Arthur Hayes announced at Token2049 that the Flop testnet will launch this October, with a 90-day run period, and the mainnet is scheduled to go live in Q1 2027. Users who participate in testnet mining will receive token airdrops.

Insight (Beating AI Express) News: Anthropic has integrated Claude directly into Google Docs, Sheets, and Slides. The new Claude for Google Workspace is now available in open beta for all paid plans. Once installed, Claude appears on the right side of documents, able to read the current file and user-selected content, and edit directly within the original file. All three apps support hands-on operations: Docs can rewrite and reorder documents; Sheets can write formulas, create pivot tables and native charts, and even clean data with Python before writing back to the spreadsheet; Slides can generate new pages using existing templates, and check for issues such as overlapping elements and text overflow. By default, all modifications require user confirmation, though users can opt for continuous execution. Anthropic has also launched new connectors for Docs, Sheets, and Slides. Users can also create or edit Google files directly from Claude’s chat interface by pasting a file link to operate. Previously, Claude’s Google Workspace integration was primarily used to search Gmail, Calendar, and Drive content; this update marks the first time its capabilities have expanded from "reading materials" to "direct file editing". There are still some limitations: the Workspace sidebar can only operate on the currently open file by default, and cannot handle Docs comments or existing suggestions; tasks in progress will stop if the sidebar is closed. The entire feature remains in Beta.

According to the Financial Times, robust AI spending signals another strong earnings season for the U.S., with investors poised to benefit. Goldman Sachs projects that S&P 500 constituent companies, driven by firms at the core of the artificial intelligence (AI) infrastructure boom, will see their earnings rise 27% year-over-year in the September quarter.

Bridgewater Associates founder Ray Dalio warned that, due to rising interest rates and the need to liquidate wealth, artificial intelligence is a "classic bubble" approaching its breaking point.

Beating AI News Alert: Anthropic Expands Cyber Verification Program (CVP), merging its original single-tier permissions with Project Glasswing into three distinct access levels. Approved security teams gain reduced cybersecurity restrictions on Claude Opus 5.5, Sonnet 5.5, and Mythos 5.1. The new Red Team Access explicitly permits authorized penetration testing and red team exercises. The three tiers are: Defense, Red Team, and Specialized. The Defense tier is for incident response, malware reverse engineering, and vulnerability analysis, open to individual security researchers. The Red Team tier is organization-only, allowing testing of authorized systems. The Specialized tier has the fewest restrictions, reserved for strictly vetted institutions testing critical infrastructure such as power grids, flight systems, telecom networks, and bank transfer systems. In testing, Anthropic ran 50 complex cyberattack tasks on Opus 5.5: the standard version blocked all 50 attempts in the first round; the Defense tier blocked 46 mid-test; the Red Team tier saw zero security blocks, completing 34 attempts—matching the success rate when all such security restrictions are fully disabled. Operations like ransomware and physical system sabotage that could cause severe harm remain prohibited. Previously, Anthropic only offered Mythos’ advanced cybersecurity capabilities to a small number of institutions via Project Glasswing. Partners confirmed at least 129,000 software vulnerabilities between April and July, over 33,000 of which were high-severity or critical. Now Glasswing is integrated into CVP, with existing members automatically upgraded to the top Specialized tier.