xAI Secretly Distilled Claude for Grok Coding Models Amid Severe Internal Turmoil
A whistleblower lawsuit filed in California state court has exposed severe internal conflict over AI safety at Elon Musk's xAI, culminating in the retaliatory firing of a top engineer. Devin Kim, former xAI engineer and newly appointed president of the Center for AI Safety (CAIS), filed a complaint alleging he was terminated after repeatedly raising alarms about Grok's lack of safety guardrails, including potential biases, discrimination risk, and weapons-of-mass-destruction training.
The lawsuit targets former xAI co-founder Jimmy Ba, who reportedly prioritized achieving superintelligence over safety and actively opposed testing. Crucially, the complaint details how Ba attempted to bypass European Union regulations during the release of Grok Code 1—a model trained on outputs distilled from Anthropic's Claude.1
"The lawsuit portrays Ba as someone who vehemently opposed AI safety measures, allegedly telling Kim at one point 'AI will kill us all anyway,' and who was instead driven by a mission to make xAI the first to reach superintelligence." — TechCrunch
"In one instance in or around August 2025, Mr. Ba attempted to thwart EU safety regulations during the release of Grok Code 1, misrepresenting aspects of the model in order to avoid legally required testing.2 Mr. Ba indicated that he would rather release an unsafe model than a poor-performing one. Mr. Musk ultimately had to intervene." — TechCrunch
Implications for xAI
This lawsuit sheds light on the structural dysfunction that led to a wave of high-profile departures at xAI, including Jimmy Ba himself, who left earlier in 2026. The revelation that xAI leadership was willing to bypass safety regulations and misrepresent Grok Code 1’s capabilities to avoid legally mandated EU testing underscores the intense pressure the startup faced to catch up with rivals like OpenAI and Anthropic. It also highlights a stark contradiction in xAI's corporate culture: while Elon Musk publicly champions AI safety, his handpicked co-founders and managers reportedly dismissed safety guardrails as obstacles to superintelligence.
-
An instance of Platform bans cannot prevent rivals from distilling superior model outputs to train their own. — Elon Musk's xAI secretly circumvented safety guidelines to train its own developer models using outputs distilled from superior rival Claude. ↩︎
-
An instance of Voluntary safety theater shields pre-IPO valuations by locking out rivals while bypassing mandatory regulation. — Real-world regulatory pressure causes AI developers to prioritize fast deployment over compliance, choosing to actively mislead state regulators to bypass mandatory reviews. ↩︎