Anthropic Previews Claude Duet, ChatGPT Adds PIN-Locked Chats, and Texas Sues OpenAI Over ChatGPT Safety

Three stories from the past few days point at the same tension: assistants are becoming more proactive and more personal, while the people who regulate them are getting more aggressive. Details come from press coverage of vendor announcements and court filings, so treat product behavior as the vendors' own description and legal claims as allegations.

1. Anthropic Previews Claude Duet, an Agent That Doesn't Wait to Be Asked

On October 21, Anthropic announced Claude Duet, a research preview of a proactive personal agent built on Claude Opus 5.5. It runs as a separate app on iOS, Android and the web, and access is a waitlist for U.S. Claude Pro and Max subscribers. Per press coverage, it sends a short morning brief on important items from connected apps, then re-checks those apps every few hours and alerts you when something needs attention.

Two design choices stand out. First, it asks for approval before consequential actions such as sending a message, making a payment, signing a document or deleting a file. Second, background checks count against your existing plan limits, so proactivity has a visible cost. Anthropic has acknowledged the agent can still misjudge what is urgent, and I could not find Anthropic's own announcement text, so these details rest on secondary reports.

For builders, the shift is from request-response to scheduled, event-driven loops. The hard problem is no longer answering well, it is deciding when to interrupt.

2. ChatGPT Adds Locked Chats

On October 20, OpenAI began rolling out Locked Chats: conversations sealed behind a four-digit PIN, Face ID or Touch ID, moved to their own section and excluded from history, pins and search. Reports say locked chats get a separate memory that does not feed regular Memory. They are not end-to-end encrypted, do not change retention or training settings, and cannot be applied to temporary, group, shared-link or scheduled-task chats. Resetting the PIN permanently deletes everything locked behind it.

The lesson is about honest framing. This is access control on the device and in the UI, not a cryptographic vault, and the trade-offs (separate memory, destructive reset) are product decisions your own app will face if you add a "private" mode.

3. Texas Becomes the Latest State to Sue OpenAI

On October 23, Texas Attorney General Ken Paxton sued OpenAI and Sam Altman in Collin County district court, citing the state's consumer protection law and the SCOPE Act and seeking over $1 million in monetary relief. It follows suits filed on October 14 by eight other states, which allege the chatbot harmed children and deceived adults. These are allegations at the complaint stage, with no rulings reported, and coverage notes the Texas filing comes during Paxton's U.S. Senate campaign.

Whatever the outcome, the pattern is clear: consumer-protection and child-safety statutes are now being applied to general chatbots, not just companion apps.

What Connects the Three

A proactive agent needs approval gates, a private mode needs honest limits, and both need to survive scrutiny. A minimal pattern for the first:

const needsApproval = new Set(['send_message', 'payment', 'sign', 'delete']);

async function runTool(call: ToolCall, user: User) {
  if (needsApproval.has(call.name)) {
    await queueForApproval(user, call); // user confirms in-app, then we execute
    return { status: 'pending_approval' };
  }
  return execute(call);
}

A simple flow diagram of "background check, brief, alert, approval" would make a good visual for a follow-up.

Conclusion

Proactive agents, private modes and legal exposure are one design problem: decide what the assistant may do unprompted, what it must ask first, and what you can truthfully tell users about their data.

— Maya