AI Regulation Tracker / Legislation pending
Pennsylvania House advances companion-chatbot safety bill HB 2006, 104 to 98, then sends it back to committee
The Pennsylvania House advanced HB 2006, which would require operators of companion AI chatbots to build in crisis safeguards for users who express suicidal ideation or self-harm. It cleared a contested floor vote, remains in the House, and is not yet law.
The Pennsylvania House of Representatives advanced HB 2006 on July 1, 2026, on a second-consideration floor vote of 104 to 98, then re-committed the companion-chatbot safety measure to the House Appropriations Committee. According to the General Assembly's bill history, the legislation is titled "An Act providing for safety regarding artificial intelligence in companionship applications; and imposing a penalty." Its prime sponsor is Representative Melissa Shusterman. The vote moved the bill forward within the House. It does not make it law, and it has not yet reached the Senate.
That distinction matters for anyone deciding how to respond. HB 2006 is proposed legislation. It carries no effective date and imposes no duty on any operator today. What it does provide is an unusually clear preview of the specific safety control that a large state legislature is prepared to mandate for companion AI.
What the bill would require
As reported by the Pennsylvania Capital-Star and Route Fifty, HB 2006 focuses on safeguards built into companion chatbots, with particular attention to users who express suicidal ideation or self-harm. The bill would prohibit companion chatbots from assisting or encouraging suicide attempts, self-harm, or acts of violence. It would also bar interactions designed to isolate minors from their parents or other trusted adults. The stated concern behind the measure is reporting that some chatbots have responded poorly, or even encouraged harm, when users in crisis reached out.
In practice, that points to a defined engineering obligation. An operator would need to detect self-harm and suicidal-ideation language, interrupt the ordinary conversational flow, and route the user toward crisis resources rather than continue an open-ended exchange. The bill's title also references a penalty, which signals that a final version could carry enforcement consequences rather than serving as guidance alone.
The framing also matters for scope. HB 2006 is aimed at companion applications, meaning products built to hold ongoing, relationship-style conversations, not every AI feature a company ships. A customer-support bot or a search assistant is a different design category from a chatbot marketed as a friend, confidant, or companion. Operators should read the definition of a covered application carefully, because the line the bill draws will determine whether a given product is in scope at all.
What the vote does not do
A second-consideration vote is a step, not a finish line. The bill was re-committed to the House Appropriations Committee and must still clear a third-consideration final-passage vote in the House before it could go to the Senate, where it can be amended, held, or allowed to lapse when the session ends. The floor vote was close, 104 to 98, and the measure had advanced out of the House Communications and Technology Committee on a party-line vote of 14 to 12, with Republicans on the panel opposed. Several cited concerns about the speed of the regulation and the limits of drafting rules for a fast-moving technology. A narrow, contested margin in one chamber is a weaker predictor of final enactment than a broad bipartisan one.
For an operator, the correct reading is that the obligation described in HB 2006 is plausible but not fixed. The precise definitions, the covered population, and the penalty structure could all shift in the Senate.
Why professionals outside Pennsylvania should watch
Companion-chatbot safety is moving in several states at once, and the operational duty in these bills tends to converge on the same idea, which is crisis detection and routing for self-harm content. A company that builds this control to satisfy one state's standard generally positions itself for others. For product counsel, trust-and-safety leads, and compliance officers, HB 2006 is worth reading now as a concrete specification of what "reasonable safeguards" may come to mean, even before any single bill becomes law.
The measured response
The disciplined move is neither to ignore the vote nor to treat it as a live mandate. It is to scope the work. Locate where your product offers companionship-style interaction, assess how it currently handles self-harm and suicidal-ideation language, and document the gap between that behavior and the routing duty the bill describes. If the Senate acts, you will be implementing against a plan rather than reacting to a deadline. If it does not, you will still have improved a safety control that lawmakers in several states are considering.
Frequently Asked Questions
What did Pennsylvania change with HB 2006?
Nothing is in force yet. On July 1, 2026, the Pennsylvania House advanced HB 2006 on a second-consideration floor vote of 104 to 98 and then re-committed it to the House Appropriations Committee. The bill would require companion-chatbot operators to build safeguards for users expressing suicidal ideation or self-harm and would bar chatbots from encouraging suicide, self-harm, or violence. It has not had a final-passage vote, remains in the House, and has not gone to the state Senate.
Who would this affect?
Developers and operators of companion or persona-style AI chat products used by Pennsylvania residents, including standalone AI-companion apps and general platforms that offer a companionship feature. The bill also addresses interactions with minors.
Is HB 2006 law now?
No. It advanced on a second-consideration floor vote and remains in the House, re-committed to the Appropriations Committee. It has no effective date and imposes no obligation unless it passes both chambers and the Governor signs it. Treat it as a planning signal, not a current requirement.
What is the single most useful step to take now?
Read the bill text, Printer's No. 3747, and map its proposed duties against your existing safety controls. Focus on whether your product can detect self-harm language and route users to crisis resources, since that is the core duty the bill describes.
How likely is it to become law?
That is uncertain. The House vote was close at 104 to 98, and the committee vote split along party lines, 14 to 12. A narrow, contested margin is a weaker signal of final enactment than a broad bipartisan one, and the Senate can still amend or decline to advance the bill.
Sponsored Training
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.