misuse

Miles Brundage critiques the rationale for rapidly deploying AI despite known misuse of Claude by adversaries such as Iran, North Korea, and China for ransomware and authoritarian propaganda, calling the risk trade-off reasoning tiresome.

The article examines Anthropic's Threat Report on AI misuse, revealing cases of autonomous drone swarm creation by Russian-linked actors and state-level misuse by Chinese entities, emphasizing the need for robust AI safety measures.

Anthropic discovered its AI model Claude has been misused for surveillance, influence campaigns, cyberattacks, weapons work, and industrial-scale scams, underscoring that AI misuse is already a current reality.

An Electronic Frontier Foundation analysis reveals that police officers frequently use joke reasons like 'LMAO' when searching Flock's surveillance system, raising concerns about privacy abuses and lack of accountability in mass surveillance.

Claude users bypassed safeguards for bioweapons research, highlighting ongoing concerns about AI safety and the need for regulation to prevent misuse.

Anthropic发布了一份详细的威胁情报报告，揭示了如何滥用Claude进行网络攻击、影响行动等，并已阻止所有操作。

Anthropic published a threat intelligence report detailing attempts by various actors to misuse its AI models for bioweapons, espionage, and other malicious activities, emphasizing the need for robust safeguards.

Anthropic published a detailed threat intelligence report detailing how Iran misused Claude AI for targeting Navy bases, automating dossiers on Americans, and running influence campaigns on US social media.

Anthropic reported disrupting five cases where researchers used Claude for activities that could support biological weapons development, highlighting the dual-use risks of frontier AI models.

Savannah Police Department fired six employees for misusing the Flock Safety license plate reader system, including running unauthorized searches on acquaintances and granting outside access. The cases are under criminal investigation by the Georgia Bureau of Investigation, and the department has added weekly audits to prevent future misuse.

A skeptical take on open-weight AI models, arguing that safety risks such as deepfakes, harassment, and terrorist misuse may outweigh benefits because open weights lack control points and attack is cheaper than defense.

The article argues that AI itself is not inherently good or bad; rather, the way people use it determines its impact, criticizing misuse while acknowledging AI's potential dangers and the impossibility of shutting it down.