Anthropic researchers developed a method using large language model agents to evaluate social science coding tasks, finding that they not only match human coding accuracy but help surface ambiguous cases, make reasoning explicit, and reduce costs and time. The approach shows promise for scalable, transparent content analysis in social science research.
Anthropic researchers developed a method using large language model agents to evaluate social science coding tasks, finding that they not only match human coding accuracy but help surface ambiguous cases, make reasoning explicit, and reduce costs and time. The approach shows promise for scalable, transparent content analysis in social science research.
Protestware is code intentionally introduced by developers for political statements. AI coding agents that automatically pull and execute code may be particularly vulnerable to such protestware, posing security and ethical risks for automated development pipelines.
The article warns that "protestware" — code injected into projects for political protest — could be used to target AI-powered coding agents. As these tools automatically fetch and execute code from open-source dependencies, they may unknowingly run malicious or protest-oriented payloads, raising new security and ethical concerns.
The article discusses the concept of "protestware" designed specifically for coding agents (AI-powered coding assistants), exploring how developers could embed political or ethical protest messages into code that would be triggered when AI agents process or generate software.
No deep-dive for this story yet — use the button below to generate one.
Anthropic researchers developed a method using large language model agents to evaluate social science coding tasks, finding that they not only match human coding accuracy but help surface ambiguous cases, make reasoning explicit, and reduce costs and time. The approach shows promise for scalable, transparent content analysis in social science research.