The popular AI framework Virtuals Protocol will host an in-person meet-up in Hong Kong on Friday, January 17.

Insight Beating AI Flash News: Anthropic announced it has brought six researchers from independent AI evaluation firm METR in-house, providing them with office space, work computers, and system permissions roughly matching those of its internal security assessment team. No longer just testing Claude from outside, the researchers now access internal tools to probe the model’s dangerous behaviors, security incidents, and protective safeguards. Anthropic compared this setup to permanent inspectors deployed by bank regulators to financial institutions. METR may publicly release its risk assessments and investigation findings independently, with Anthropic having no authority to block such reports. However, for customer data, legally confidential materials, and model architecture, Anthropic can still demand removal of relevant content. If METR considers the deletion unreasonable, it may publicly voice objections but cannot disclose the redacted original text. Anthropic categorizes external evaluations into three tiers, with METR now holding the highest access level. Another non-profit, SecureBio, is reviewing its bioweapon risks, while Accenture’s Faculty is also conducting model red teaming tests and security assessments. METR explicitly states it does not accept compensation or donations from frontier AI companies, while Anthropic previously announced it would directly fund Accenture’s evaluation work. In an October 14 statement, METR acknowledged that the current cooperation still relies on AI firms’ voluntary information provision, meaning evaluators may lack access to key evidence, and vendors can shape public visibility by restricting access and redacting reports.

Machi(@machibigbrother) generated $1.41B in trading volume on @papertrade_xyz and lost $416K. But he earned 57.6M $PAPER tokens and received $341K in USDC rewards by staking. This brings his total loss down to $74.7K.

SlowMist has issued a security alert warning of a counterfeit "Ledger firmware update" campaign exploiting recent Ledger supply chain incidents to defraud users. The fake website replicates processes including "authenticity verification", device selection, and "Update Now", with a highly realistic interface. The alert advises users not to update their devices via website links, and never to input their recovery phrases anywhere.

Upbit has added HBD and DNT to its transaction warning asset list, and suspended deposit and withdrawal services for the two tokens.

Per Lookonchain monitoring, "Big Brother Maji" (Huang Licheng) executed $1.41 billion in trades on the Papertrade platform, suffering a $416,000 loss. He received 57.6 million PAPER tokens and $341,000 worth of USDC rewards via staking, bringing his total loss down to $74,700.

According to TradingBeats' monitoring, contracts for Micron, SK Hynix, and SanDisk on Hyperliquid have all declined notably from their respective peaks since the late-July rebound. However, their relative strength during that prior rebound phase varied: Micron is currently the most resilient to drops, while SK Hynix has recorded the largest pullback. Micron (MU) has fallen from its October 2 high of $1,114.90 to $923.62, a total pullback of roughly 17.16%—the smallest among the three. SanDisk (SNDK) has dropped from its September 22 high of $1,908 to $1,378.40, a total pullback of around 27.76%, roughly 10.60 percentage points more than Micron. SK Hynix (SKHX) has fallen from its September 10 high of $1,438.40 to $958.73, a total pullback of around 33.35%—the largest among the three, underperforming Micron by approximately 16.19 percentage points. During the earlier rebound period, however, the top performer was SanDisk, which now sits in the middle of the three in terms of pullbacks, while Micron posted the smallest gains. Using the July 31 intraday lows of each asset as the storage sector's trough, SanDisk rallied to a peak of roughly 60.27%, SK Hynix to around 38.73%, and Micron to about 36.14%.