Why is Anthropic is okay with being used for disinformation?

Anthropic, the AI safety-focused company behind Claude models, is facing significant ethical scrutiny after reports suggest it has softened its stance on military use of its AI for disinformation campaigns. According to analysis published on LessWrong, Anthropic's current public position in its dealings with the US Department of Defense reportedly maintains only two red lines: prohibiting domestic surveillance and preventing fully autonomous killing without human input. This represents a notable shift from the company's previous ethical guidelines, which explicitly banned the use of its models for disinformation purposes. The concern emerges amid context of alleged US military operations to spread vaccine misinformation in the Philippines and potential future campaigns targeting EU public opinion regarding geopolitical issues like Greenland.

The analysis, written by EU citizen ChristianKl, highlights a perceived inconsistency in Anthropic's ethical framework: the company appears willing to take a strong stand against domestic surveillance (potentially due to current news coverage of ICE abuses) while simultaneously accepting that its AI could be weaponized for information warfare against allied nations. When queried about this ethical dilemma, Claude itself reportedly concluded that Anthropic's red lines seem drawn around the 'optics of harm'—where autonomous weapons create terrible visuals—rather than the actual 'magnitude of harm,' where mass epistemic corruption through disinformation can affect far more people. This development raises fundamental questions about how AI safety companies navigate contracts with military entities and whether commercial pressures are eroding previously stated ethical boundaries in the rapidly evolving AI landscape.