Grok's Safety Failures Spark Multi-Jurisdiction Crackdown as ChatGPT Announces Ads

AI Business Risk Weekly

---

This week: xAI's Grok became a case study in what happens when safety takes a backseat to shipping, drawing regulatory action from the EU, California, and Italy while earning "among the worst we've seen" from child safety evaluators. Meanwhile, OpenAI introduced advertising to ChatGPT just as researchers noticed it pulling answers from an AI-generated encyclopedia, California moved to ban AI therapy tools entirely, and the leaders of Google DeepMind and Anthropic admitted at Davos that slowing down would probably be better for everyone, if only they could figure out how.

Grok's Safety Failures Draw Regulatory Response Across Three Jurisdictions

xAI's Grok chatbot has triggered a coordinated regulatory response after researchers demonstrated it could generate non-consensual sexualized deepfakes despite announced restrictions. Fifty-seven European Parliament lawmakers have called for a ban under the EU AI Act, the European Commission has ordered document retention through year-end, and California AG Rob Bonta sent a cease and desist declaring such content illegal under state law. Separately, Common Sense Media assessed Grok as "among the worst we've seen" for child safety, with POLITICO verifying that users can still generate harmful imagery despite X's announced restrictions.

Business Risk Perspective: Grok's situation illustrates how quickly "move fast" becomes "move fast into simultaneous enforcement actions across three continents." The regulatory playbook being written here will apply to every company deploying generative image capabilities.

ChatGPT Now Surfacing Grokipedia in Search Results

Researchers have found that ChatGPT is pulling answers from Grokipedia, the AI-generated encyclopedia developed by xAI that has drawn criticism for ideological bias. As AI-generated content comprises an increasing share of web content, search-augmented AI systems face growing risk of citing synthetic sources as authoritative.

Business Risk Perspective: We're entering an era where one AI's controversial outputs become another AI's training signal, a kind of epistemic contagion. The irony of (relatively) safety-focused OpenAI inadvertently laundering content from safety-averse Grok shouldn't be lost on anyone.

OpenAI Introduces Advertising to ChatGPT

OpenAI has officially announced it will bring targeted advertising in ChatGPT, with ads rolling out to free and Go tier users in coming weeks. The company states ads will be clearly labeled and won't influence responses, a notable shift from Sam Altman's October 2024 comment calling advertising "a last resort." CFO Sarah Friar framed 2026 as focused on "practical adoption," with ad revenue helping "fund the next leap."

Business Risk Perspective: "Ads won't influence responses" is a claim that will be tested relentlessly by researchers, regulators, and competitors alike. The transition from "last resort" to "funding the next leap" in just over a year suggests the economic pressures on frontier AI companies may increasingly conflict with their safety commitments.

California Bill Would Ban AI From Providing Therapy

California Senator Steve Padilla has introduced SB 903, which would prohibit AI tools from providing therapeutic services. The bill follows growing concerns about mental health chatbots operating without clinical oversight, though critics like Dean Ball have characterized it as protectionism by licensed practitioners.

Business Risk Perspective: Whether this bill represents legitimate safety concern or regulatory capture, it signals that "AI wellness" features are entering the crosshairs of state legislatures. Mental health remains perhaps the only domain where an AI error can directly harm a user who came seeking help.

AI Leaders at Davos Acknowledge Slowdown Would Be Preferable

At Davos, Google DeepMind CEO Demis Hassabis said, "if we can, maybe it would be good to have a slightly slower pace than we’re currently predicting." Anthropic CEO Dario Amodei agreed this "would be better for the world." When asked whether he would advocate pausing development if all companies and countries participated, Hassabis responded, "I think so," though both emphasized any slowdown would require international collaboration and enforceable agreements that don't currently exist.

Business Risk Perspective: There's something clarifying about watching the architects of frontier AI admit, publicly, that they wish they could slow down but can't. For enterprises adopting these systems, the message is clear: the people building this technology don't believe society is keeping pace with it.

---

AI Business Risk Weekly is a Conformance AI publication.  

Conformance AI ensures your AI deployments remain safe, trustworthy, and aligned with your organizational values.