Anthropic takes aim at chatbot ads—with its own Super Bowl ad

Welcome to AI Decoded, Fast Company’s weekly newsletter that breaks down the most important news in the world of AI. You can sign up to receive this newsletter every week via email here.

Anthropic uses the Super Bowl to land some zingers about the future of AI

Anthropic’s Super Bowl ads are bangers. The spots, which Anthropic posted on X on Wednesday, seize on rival OpenAI’s plans to begin injecting ads into its ChatGPT chatbot for free-tier users as soon as this month. The 30-second ads dramatize what the real effects of that decision might look like for users. They never mention OpenAI or ChatGPT by name.

In one ad, a human fitness instructor playing the role of a friendly chatbot says he’ll develop a plan to give his client the six-pack abs he wants, before suddenly suggesting that “Step Boost Max” shoe inserts might be part of the solution. In another, a psychiatrist offers her young male patient some reasonable, if generic, advice on how to better communicate with his mom, then abruptly pitches him on signing up for “Golden Encounters,” the dating site where “sensitive cubs meet roaring cougars.”

The ads are funny and biting. The point, of course, is that because people use chatbots for deeply personal and consequential things, they need to trust that the answers they’re getting aren’t being shaped by a desire to please advertisers.

OpenAI CEO Sam Altman, however, was not laughing. He responded to the ads by saying his company would never run ads like the ones portrayed by Anthropic. But he didn’t stop there. He went much further. “Anthropic wants to control what people do with AI,” he wrote in a long post on X on Wednesday. “They block companies they don’t like from using their coding product (including us), they want to write the rules themselves for what people can and can’t use AI for, and now they also want to tell other companies what their business models can be.” He went on to call Anthropic an “authoritarian company.”

Anthropic, which makes its money through subscriptions and enterprise API fees, says it wants its Claude chatbot to remain a neutral tool for thinking and creating. “[O]pen a notebook, pick up a well-crafted tool, or stand in front of a clean chalkboard, and there are no ads in sight,” the company said in a blog post this week. “We think Claude should work the same way.”

By framing conversations with Claude as a “space to think” rather than a venue for ads, the company is using the Super Bowl’s massive cultural platform to question whether consumer marketing is the inevitable future of AI.

How social media lawsuits could affect AI chatbots

AI developers (and their lawyers) are closely watching a long-awaited social media addiction trial that recently kicked off in a Los Angeles courtroom. The case centers on a 20-year-old woman who alleges that platforms including Facebook and Instagram used addictive interface designs that caused her mental health problems as a minor. The suit is part of a joint proceeding involving roughly 1,600 plaintiffs accusing major tech companies of harming children. TikTok and Snap have already settled with plaintiffs, while Meta and YouTube remain the primary defendants.

While Meta has never admitted wrongdoing, internal studies, leaked documents, and unsealed court filings have repeatedly shown that Instagram uses design features associated with compulsive or addictive engagement, and that company researchers were aware of the risks to users, especially teens.

What makes the case particularly significant for the AI industry is the legal strategy behind it. Rather than suing over content, plaintiffs argue that the addictive features of recommendation algorithms constitute harmful product defects under liability law. AI chatbots share key similarities with social media platforms: they aggregate and dispense content in compelling ways and depend on monetizing user engagement. Social networks rely on complex recommendation systems to keep users scrolling and viewing ads, while AI chatbots could be seen as using a different kind of algorithm to continually deliver the right words and images to keep users prompting and chatting.

If plaintiffs succeed against Meta and YouTube, future litigants may attempt similar “addictive design” arguments against AI chatbot makers. In that context, Anthropic’s decision to exclude ads—and to publicly emphasize that choice—may help it defend itself by portraying Claude as a neutral, utilitarian tool rather than an engagement-driven “attention trap.”

No, OpenClaw doesn’t herald the arrival of sentient AI agents

Some hobbyists and journalists have gone into freakout mode after seeing or using a new AI agent called OpenClaw, formerly Clawdbot and later Moltbot. Released in November 2025, OpenClaw is an open-source autonomous AI assistant that runs locally on a user’s device. It integrates with messaging platforms like WhatsApp and Telegram to automate tasks such as calendar management and research. OpenClaw can also access and analyze email, and even make phone calls on a user’s behalf through an integration with Twilio. Because personal data never leaves the user’s device, users may feel more comfortable giving the agent greater latitude to act autonomously on more complex tasks.

One user, vibe-coding guru Alex Finn, posted a video on X of an incoming call from his AI agent. When he answered, the agent, speaking in a flat-sounding voice, asked whether any tasks were needed. Finn then asked the agent to pull up the top five YouTube videos about OpenClaw on his desktop computer and watched as the videos appeared on screen.

Things grew stranger when AI agents, including OpenClaw agents, began convening on their own online discussion forum called Moltbook. There, the agents discuss tasks and best practices, but also complain about their owners, draft manifestos, and upvote each other’s comments in threaded “submolts.” They even generated a concept album, AVALON: Between Worlds, about the identity of machines.

That behavior led some observers to conclude that the agents possess some kind of internal life. Experts were quick to clarify, however, that this is a mechanical illusion created by clever engineering. The appearance of “independence” arises because the agents are programmed to trigger reasoning cycles even when no human is prompting them or watching. Some of the more extreme behaviors, such as “rebellion” manifestos on Moltbook, were likely prompted into existence by humans, either as a joke or to generate buzz.

All of this has unfolded as the industry begins to move from the “chatbot” phase into the “agent” phase of generative AI. But the kinds of free-roaming, autonomous behaviors on display with OpenClaw are not how the largest AI companies are approaching the shift. Companies such as Google, OpenAI, and Anthropic are moving far more cautiously, avoiding splashy personal agents like “Samantha” in the movie Her and instead gradually evolving their existing chatbots toward more limited, task-specific autonomy.

In some cases, AI labs have embedded their most autonomous agent-like behaviors in AI coding tools, such as Anthropic’s Claude Code and OpenAI’s Codex. The companies have increasingly emphasized that these tools are useful for a broad range of work tasks, not just coding. For now, OpenAI is sticking with the Codex brand, while Anthropic has recently launched a streamlined version of Claude Code called CoWork, aimed at general workplace tasks.

More AI coverage from Fast Company: 

• AI can now fake the videos we trust most. Here’s how to tell the difference
• Moltbook, the viral social network for AI agents, has a major security problem
• AI in healthcare is entering a new era of accountability
• What happens to the AI exit market if the FTC cracks down on ‘acquihires’?

Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company Premium.

View the full article