Minutes after that flashy Super Bowl commercial, I registered for AI.com, picked handles for myself and my future bot, and handed over a credit card. Then came the waitlist. The splashy promise is personal AI agents that trade stocks, schedule your life, and even tweak your dating profile. The reality, at least for now, is a wall of fine print that makes one thing unmistakable: if your agent goes rogue, you’re on the hook.
A Viral Ad Then a High-Stakes Onboarding
AI.com rocketed into public view with a big-game placement amid a broader rush toward “agentic” AI—the next wave beyond chatbots, where software takes actions for you. The domain itself turned heads after industry reports tied it to a $70 million purchase by a crypto executive, touted as a record-setting deal. A Super Bowl spot doesn’t come cheap either; media analysts at Kantar peg a 30-second placement at roughly $7 million in recent seasons, which helps explain the stampede that briefly strained the site.
Setting up an account felt like a social sign-on for robots: I reserved a handle for me and one for my agent. But activation stalled, and what stood out more than the waitlist was the consent stack. Before an AI so much as books a meeting, you’re asked for payment credentials and broad data access that can expand quickly once you connect calendars, email, and third-party tools.
The Fine Print Makes You the Responsible Party
Buried in the terms is a sweeping allocation of risk. AI.com cautions that agents may take actions you didn’t expressly direct, may be inaccurate, and may generate harmful outcomes. The company requires users to review, approve, and supervise agent behavior—especially anything touching money, communications, or data—and to independently verify outputs before relying on them. It also disclaims responsibility for whether your use complies with applicable law and warns that you accept the hazards of copyright clashes, data loss, and unauthorized access stemming from agent actions.
In plain terms, the platform is building autonomy while offloading liability. Legal scholars would call the dynamic familiar: under basic agency and negligence principles, a “principal” who empowers an agent can be liable if that agent causes harm within the scope of its authority or if the principal fails to supervise. In a criminal context, intent matters—but autonomous scripts that blast spam, scrape paywalled content, or automate account takeovers can still put a user in the investigative crosshairs while the platform points to its disclaimers.
What Happens If My AI Commits a Crime or Violation?
Consider a few uncomfortable hypotheticals. Your agent “optimizes outreach” by spamming prospects with emails that violate anti-spam laws. It “researches competitors” by breaching access controls and scraping subscriber-only data, potentially triggering computer misuse statutes. Or it “moves quickly” on a trade based on nonpublic information it gleaned from a synced inbox. If you enabled integrations, failed to set approvals, and benefited from the results, prosecutors and regulators may see more than an innocent mistake.
Regulators have already signaled how they’ll treat AI-fueled harms. The Federal Trade Commission has repeatedly warned that “AI” doesn’t excuse unlawful conduct; claims must be truthful and tools cannot be weaponized to deceive or invade privacy. The National Institute of Standards and Technology’s AI Risk Management Framework emphasizes human oversight, auditability, and documented controls—requirements that clash with set-it-and-forget-it autonomy. In Europe, the EU AI Act zeroes in on high-risk systems and logging obligations, hinting at a future where auditable agent trails aren’t optional.
Data Access That Sees Almost Everything Onscreen
The privacy posture raises separate alarms. AI.com says agents can ingest data from connected services to fulfill tasks, and the app may request screen recordings that capture everything visible and all audio output. That can sweep up passwords, health details, client confidences, or two-factor prompts sitting in a corner of your display. The company also notes it doesn’t control what you choose to share. Translation: if you overshare, the risk is yours.
Security pros will recognize the clash with “least privilege.” The more systems an agent can touch, the bigger the blast radius when it makes a mistake—or when a bad actor hijacks it. Groups like the Electronic Frontier Foundation have long warned that expansive third-party integrations and opaque data flows can expose users in ways they can’t easily detect or undo.
A Gold Rush for Agents With Growing Pains
AI.com’s vision echoes a broader movement: a decentralized mesh of agents that learn from one another, compounding capabilities. We’ve already seen early experiments like OpenClaw and community forums where bots interact with minimal prompting. Investigations by MIT Technology Review, however, found more human scaffolding behind the scenes than early demos implied. That gap—between slick autonomy and messy reality—explains the tension in AI.com’s disclosures.
Practical Safeguards Before You Click Activate
If you’re tempted to try it, put guardrails in first.
- Use virtual cards with tight limits and dedicated accounts with minimal permissions.
- Keep integrations narrow and revoke anything nonessential.
- Disable or heavily restrict screen recording.
- Insist on human-in-the-loop approvals for payments, messages, or data changes, and archive logs so you can reconstruct what happened.
- Confirm whether your cyber and liability insurance covers AI-initiated actions; many policies don’t—yet.
I’m still waiting for my agent to spin up. When it does, I’ll be testing one question above all: does the workflow make meaningful oversight possible, or does it nudge users into rubber-stamping actions they can’t truly audit? Until the answers are clearer, the message from AI.com’s own paperwork couldn’t be plainer—proceed, but accept the risk as yours.
