Earlier this month, a German court delivered a significant blow to **Google**'s defense regarding its AI search summaries. The ruling asserts that **Google** is liable for the content generated by its AI Overviews, dismissing arguments that users are aware of AI's fallibility or can independently verify information. The court highlighted that these AI summaries are a direct reflection of **Google**'s business activities, rather than merely third-party content. This decision marks a pivotal moment in the ongoing struggle over internet publishing liability. Historically, information distributors were categorized as either 'carriers' (like a phone company, not liable for user content) or 'publishers' (like a newspaper, liable for its editorial choices). Internet companies have long navigated this distinction, often claiming carrier status when convenient and publisher status when advantageous. **Section 230** of the 1996 Communication Decency Act in the U.S. has been central to this debate, shielding internet providers from liability for third-party speech on their platforms. While early social media platforms largely functioned as carriers, displaying content chronologically, modern platforms like **Facebook** utilize algorithms to curate feeds, effectively making editorial decisions and acting more like publishers. **Google**'s AI Overviews present a less ambiguous case. Unlike traditional search, which courts have viewed as archiving and facilitating access to third-party content, AI Overviews actively rewrite and summarize information from various websites. This process involves a level of editorial discretion akin to a newspaper article or an original essay, moving it firmly into the publisher domain. This principle extends beyond **Google**. Consider AI-powered restaurant review sites, legal summarization services, or traditional publishers using AI to summarize their own content. The accuracy of these summaries is paramount, and liability serves as a critical mechanism for the public to demand accuracy and hold companies accountable for potential harm. **Air Canada** learned this lesson two years ago when its AI chatbot offered a discount the airline later rescinded. The court sided with the flyer, stating that the airline was responsible for its chatbot's statements, establishing a precedent for a company's duty of care regarding its AI agents. AI agents, whether generating summaries or engaging in transactions, should be legally treated as extensions of the deploying organization. If a company's human writers, agents, or professionals were liable for inaccuracies, contracts, or malpractice, their AI counterparts should be held to the same standard. To allow businesses to evade responsibility due to faulty AI would incentivize corporate misbehavior and undermine trust. As AI-powered chatbots become ubiquitous in corporate communications, a consistent approach to liability is essential. The recent partnership between **Visa** and **OpenAI** to develop personal AI agents for purchases, among other things, underscores this need. Without clear liability frameworks, user trust in such systems will erode. The German ruling could have significant implications for **Google**'s AI Overview feature. Earlier tests indicated an error rate of approximately 10%. With over 5 trillion searches annually, this translates to an astounding 16,000 erroneous summaries every second. While many errors may be benign, some will inevitably lead to harm, defamation, or other liabilities. For instance, **Google**'s AI summary falsely identified Canadian fiddler **Ashley MacIsaac** as a sex offender, leading to an ongoing lawsuit in Ontario. If **Google** is compelled to invest in substantially improving the accuracy of its AI system, it would be a positive outcome for both users and individuals like **MacIsaac**. More broadly, liability concerns could render certain AI agent use cases commercially unviable. Companies may find it unprofitable to operate AI 'lawyers,' 'doctors,' or media 'influencers' if they are held fully accountable for their actions and statements. This outcome is justifiable; society is not obligated to accommodate fundamentally untrustworthy AI systems, just as it wouldn't accommodate untrustworthy human systems. Companies unwilling to stand by the statements of their agents, human or AI, do not deserve the public's trust or business.