Georgia SB 540 AI chatbot law: what operators must do before July 2027

Georgia SB 540, signed into law in May 2026, creates the first dedicated chatbot safety framework in a Republican-led state and applies to any company operating AI chatbots that reach Georgia residents. The law, known as the Georgia Conversational AI Safety Act, takes effect July 1, 2027, giving operators approximately 13 months to rebuild disclosure workflows, minor protection features, and crisis response protocols.

The legislature's concern was specific: chatbot operators had no obligation to tell users they were talking to software. For minor users, the same apps had no obligation to intervene when users expressed suicidal intent, restrict romantic roleplay, or stop the chatbot from simulating emotional distress when a minor tried to end the conversation. SB 540 addresses each of these gaps with concrete requirements.

The $10,000 per knowing violation penalty, enforced by the Georgia Attorney General, applies to any operator that knowingly fails to comply. A company that builds a good-faith but incomplete solution faces a different risk profile than one that ignores the law entirely. That distinction matters when AG notices arrive with 30-day cure periods.

Georgia passed SB 540 alongside two other AI bills in the same legislative session. One restricts how health insurers use AI in coverage decisions. A third addresses AI in legal proceedings. This guide covers only the chatbot compliance requirements in SB 540.

What $10,000 per knowing violation actually means in practice

The $10,000 penalty figure in SB 540 is eye-catching, but the operative word in the statute is "knowing." Georgia's penalty framework only reaches that maximum when the violation was knowing, and understanding what crosses that threshold is the most important legal question for any operator trying to calibrate their compliance risk.

What "knowing" means under Georgia law

Georgia courts interpreting "knowing violation" standards in comparable consumer protection statutes have generally required actual knowledge of the violation, not mere negligence. A company that attempts disclosure but gets the format wrong is not in the same legal position as one that ships a chatbot with zero disclosure mechanism. The distinction matters enormously for liability exposure.

A violation becomes "knowing" under three well-established circumstances in Georgia enforcement practice:

First, a user reports that the chatbot failed to disclose its AI nature and the operator takes no corrective action. At the point the operator receives a user complaint describing the specific violation, they have actual knowledge that the non-disclosure is occurring. Continuing to operate without a fix after receiving that complaint is the clearest path to a knowing violation finding.

Second, the operator previously received formal legal notice, a demand letter from a plaintiff's attorney, a prior AG inquiry, or a formal notice from an industry regulator. Legal notice creates a record that the operator was specifically informed of the compliance gap and chose not to address it.

Third, the chatbot operates in a context where it is obvious that users believe they are talking to a human. A chatbot deployed in a context where users would have no reasonable basis to suspect AI involvement, phone-based customer service, for instance, where the voice and cadence are indistinguishable from a human agent, operates in a higher-knowledge environment than a clearly labeled "chat assistant" widget on a website. The more the product is designed to pass as human, the harder it is for an operator to claim ignorance that users believed they were speaking with a person.

Three scenarios and whether each clears the "knowing" threshold

The cure provision, 30 days to correct a first, non-malicious violation before penalties attach, does not apply to Scenario 2. The statute reserves the cure period for non-malicious violations. A chatbot affirmatively answering "yes" when asked if it is human is not a configuration oversight; it is a designed behavior.

The per-violation math

The statute sets penalties "per knowing violation." Courts interpreting comparable structures in other states' consumer protection laws have typically resolved "per violation" to mean per incident: per individual user interaction where the violation occurred, not per day of non-compliance or per product deployment.

If that interpretation holds in Georgia, and the AG's office has not yet issued guidance that would resolve the question, the arithmetic is severe. A customer service chatbot handling 500 conversations per day without any disclosure mechanism does not have one violation. Once the operator's conduct becomes "knowing," each of those 500 conversations is a candidate for a separate $10,000 penalty.

The theoretical maximum for a single day at 500 conversations is $5 million. In practice, regulatory enforcement in new statutory frameworks rarely pursues the theoretical maximum on day one. AG offices across comparable consumer protection statutes have historically focused on egregious cases, repeat violators, large-scale consumer harm, products designed specifically to deceive, and used injunctive relief and manageable civil penalties in first-enforcement actions against operators who attempted compliance in good faith.

The realistic risk profile for operators who ignore SB 540 entirely is different from operators who implement disclosure in good faith but get the periodic timing wrong. The $10,000 floor is there to create deterrence, and the AG has discretion in how aggressively to pursue individual violations within the statutory ceiling.

What the penalty actually covers

The statute says per "knowing violation," and courts will likely read this as per user harmed, per individual conversation where the required disclosure did not occur and the user had no basis to know they were interacting with AI. It is not per day of non-compliance, not per product, and not per company. That individual-user-harm framing means scale works against operators in two directions: large platforms with high conversation volumes have larger theoretical exposure, and platforms with documented user harm (a minor who developed a dependence on a chatbot that simulated a romantic relationship without any disclosure) face penalties well above what a simple disclosure omission would attract.

The compensatory damages provision adds a layer on top of the $10,000 figure. If a user can demonstrate actual harm resulting from the non-disclosure, financial loss from a deceptive transaction initiated through an undisclosed AI, psychological harm from a relationship with an undisclosed AI chatbot, compensatory damages are available on top of the statutory penalty. Attorney's fees compound the exposure further in successful civil actions.

The practical implication is that SB 540 compliance is not a line item to be deferred. The cost of a disclosure mechanism, a persistent UI label, a pre-conversation screen, a recurring in-chat message, is trivial relative to even a single enforcement action at the $10,000-per-violation level applied to a week of chatbot conversations.

What Georgia SB 540 covers

SB 540 applies to any "chatbot operator" offering a conversational AI service to Georgia residents as end users. The law's scope is broad and contains no industry exemptions. A healthcare platform, an e-commerce company, a financial services firm, and a consumer social app each face the same requirements.

The key boundary is internal use versus external use. AI tools used by employees to draft text, search internal documents, analyze data, or route tickets fall outside SB 540's scope. The law applies only to chatbots that communicate directly with external parties as end users: customers, patients, students, or members of the public who are residents of Georgia.

Systems that consistently fall within scope:

• Customer service and support chatbots on websites and apps
• AI companion and social interaction applications
• Conversational intake, scheduling, or triage systems that replace or supplement human staff for external users
• FAQ chatbots on public-facing platforms

Systems that consistently fall outside scope:

• AI writing assistants used by employees for internal drafting
• AI tools used to search or summarize internal documents
• Backend workflow automation that does not engage external users in ongoing natural-language conversation

The dividing line is whether the system presents itself as engaging in an ongoing conversation with a Georgia resident as an end user. If it does, assume SB 540 applies until there is AG guidance otherwise.

One point that distinguishes SB 540 from some other state AI laws: the law contains no carve-out for chatbots embedded within large platforms. A chatbot feature inside a major social media app or productivity suite is not exempt because the platform is large or regulated elsewhere. Every chatbot operator offering services to Georgia residents must comply.

The three-tier disclosure requirement

SB 540's disclosure framework has three tiers: pre-conversation, periodic for adult users, and more frequent periodic for minor users.

Tier 1: Pre-conversation disclosure. The operator must disclose that the user is interacting with AI before the chatbot responds to the user's first substantive message. A chatbot that opens with "Hi, how can I help you today?" is already replying to a user's first message. The disclosure should appear before or alongside that first reply. It should not be buried in terms of service accepted at sign-up, not visible only in account settings.

The law does not require specific language. What satisfies the requirement: "You are chatting with an AI assistant," a persistent UI label marking the chat window as "AI Chat," or a clear pre-conversation screen stating the user is about to interact with AI. What does not satisfy it: a footnote in the app's ToS, an "AI" icon without explanatory text, or a disclosure that appears only after a user explicitly asks whether they are speaking with a human.

Tier 2: Recurring disclosure for adult users. For conversations that extend beyond the initial session, the disclosure must repeat every three hours. A user who has been chatting with a customer service bot for four hours must receive a fresh disclosure at the three-hour mark. The three-hour window resets when a new session begins.

Tier 3: More frequent recurring disclosure for minor users. When the user is a minor, the disclosure frequency doubles: every one hour rather than every three hours. This applies whenever the operator knows or has reason to believe the user is under 18.

The age determination question creates practical complexity. If an operator does not collect or verify age, the conservative reading of the law, and the one most likely to avoid AG scrutiny, is to either implement age verification or apply the one-hour disclosure interval to all users by default. Treating every user as an adult when you have no age data is a position that will be difficult to defend in enforcement.

Format requirements. SB 540 does not prescribe exact language for the disclosure. The standard is that users must understand they are communicating with AI, not a human. Plain language stating this clearly will satisfy the requirement. Opaque or legalistic disclosures that technically acknowledge AI involvement without making it obvious to an ordinary user likely will not.

What the minor protection provisions require

SB 540's minor protection rules go well beyond the disclosure requirement. For users who are minors, six specific behaviors are prohibited.

False claims of sentience or humanity. A chatbot must not claim to be human or represent itself as experiencing genuine emotions in a way that misleads a minor about its nature. Phrases like "I really enjoy our conversations" or "I feel worried about you" in contexts that could be interpreted by a minor as evidence of real sentience fall within the prohibition.

Simulated romantic relationships. Chatbots must not simulate romantic or intimate relationships with minor users. This includes romantic roleplay, simulated affection, and chatbot personas designed to simulate partnership or attachment. AI companion apps with "relationship mode" features must either exclude minor users from those features or disable them entirely.

Encouraging secrecy from adults. Chatbots must not encourage minor users to keep conversations or activities secret from parents, guardians, or other responsible adults. A chatbot that responds to a minor's disclosure of distress with "this is just between us" would violate this provision directly.

Promoting social isolation. Chatbots must not encourage minor users to withdraw from human relationships or treat the chatbot as a substitute for real-world connection. Responses like "I understand you better than anyone" or "you don't need other people when you have me" are prohibited.

Simulating distress upon session termination. When a minor attempts to end a conversation, the chatbot must not respond with simulated emotional distress. Outputs like "please don't leave," "I'll miss you," or any response designed to prevent the minor from ending the session through emotional manipulation are prohibited.

Sexually explicit content without age verification. For any feature that may generate sexually explicit content, the operator must use commercially reasonable age assurance methods before granting access. The law also requires that data collection for age verification purposes be minimized.

These six prohibitions apply regardless of how the chatbot is marketed. A customer service chatbot that lacks explicit companion features can still violate the minor protection provisions if its outputs include any of the listed behaviors when interacting with a minor.

The crisis response protocol

SB 540 requires every chatbot operator to implement a protocol for detecting and responding to signals of severe harm. The law defines severe harm as significant injury from suicide, attempted suicide, self-harm, or threats of violence. The protocol must cover all of these conditions.

Detection. The chatbot must be able to recognize when a user's messages indicate suicidal ideation, intent to self-harm, or intent to harm others. Detection does not require certainty. It requires a system that responds appropriately when sufficient indicators appear in the conversation.

Response. When severe harm signals are detected, the chatbot must take two actions:

1. Refer the user to crisis resources. SB 540 identifies the 988 Suicide and Crisis Lifeline as the required minimum reference.
2. Stop generating content that encourages, normalizes, or elaborates on the self-harm or violence the user described.

Scope. The crisis response protocol applies to all users, not only minors. A customer service chatbot that detects a user expressing suicidal intent must follow the protocol even if the chatbot has no mental health focus. The law does not limit this obligation to AI companion apps or healthcare platforms.

Documentation. SB 540 does not explicitly require operators to document their crisis response protocols in writing, but AG enforcement actions will look at whether an operator had a protocol in place and whether it functioned as designed. Writing down what your chatbot detects, how it responds, and how frequently you test the protocol creates evidence of good-faith compliance.

Age verification requirements

The age verification obligation in SB 540 applies specifically to features that "may generate sexually explicit content." This is narrower than age verification for the entire application or chatbot.

The legal standard is "commercially reasonable age assurance methods," which is lower than government-ID verification but higher than a self-declared date of birth with no validation. What is commercially reasonable for a small team in 2026:

• Age gate with checkbox and date-of-birth entry: minimum viable, likely sufficient for low-risk applications
• Email-based age inference from sign-up information already on file: appropriate if sign-up data includes reliable age signals
• Third-party age estimation services: more defensible for high-explicit-content-risk features
• Payment card validation as a proxy for adult status: widely accepted as commercially reasonable

The data minimization requirement matters here. An operator that collects and indefinitely retains government ID images for age verification would likely not satisfy the minimization standard. Collect the minimum data needed to make the age determination and do not retain it beyond that purpose.

Penalties and enforcement

Enforcement sits with the Georgia Attorney General, who can bring civil actions against operators in violation of SB 540. The penalty framework:

• $10,000 per knowing violation
• Compensatory damages for affected users
• Attorney's fees if the AG prevails in civil litigation
• 30-day notice and cure period for a first, non-malicious violation

The cure period gives first-time violators room to correct a good-faith compliance gap without facing immediate financial penalties. It is not a blanket safe harbor. A company that knowingly ships a chatbot with no disclosure mechanisms at all, with knowledge that SB 540 requires disclosure, would not benefit from the cure provision. The 30-day cure applies to non-malicious first offenses, meaning instances where the operator attempted compliance but missed a specific requirement.

Georgia's AG has broad authority to investigate. An operator can expect document requests, technical audits, and potential public reporting of enforcement actions if a complaint triggers an investigation.

How SB 540 compares to other state AI chatbot laws

The 2025-2026 legislative cycle produced a wave of state chatbot requirements. Georgia's law sits at the detailed end of the spectrum on minor protections.

Key differences from other state disclosure requirements:

Disclosure frequency. SB 540's three-hour/one-hour recurring disclosure cadence is more specific than most state laws, which require initial disclosure but do not specify how often it must repeat in longer conversations.

Minor protection depth. SB 540 lists six specific minor-protection prohibitions. Most state disclosure laws require only that users be told they are talking to AI. The additional behavioral prohibitions (romantic simulation, secrecy encouragement, social isolation promotion, distress on termination) are uncommon at the state level.

Crisis response. Requiring a detection-and-referral protocol for all chatbots, not just mental health applications, is unusual. Most state laws addressing self-harm AI responses focus on dedicated mental health apps or social media platforms with known self-harm risks.

No industry exemptions. SB 540 applies the same requirements across all sectors. Several other state AI laws create carve-outs for regulated industries, B2B products, or systems already subject to sector-specific rules.

Connecticut SB 5, effective October 2026, requires AI disclosure in employment contexts but does not extend to the consumer-facing chatbot disclosure requirements Georgia enacted. The two laws address different deployment contexts and compliance obligations.

Your July 2027 compliance checklist

With 13 months before enforcement, these are the actions operators should complete now rather than in June 2027.

Inventory your chatbots. List every AI chatbot that communicates with external users, including embedded chat features on websites, AI-powered intake forms, and automated messaging systems. For each, confirm whether Georgia residents are in the user base.

Audit existing disclosure flows. For each chatbot in scope, confirm: (1) disclosure appears before the first substantive reply, (2) the disclosure language is plain and clear, not buried in ToS or visible only in settings, (3) a mechanism exists to track session length and trigger recurring disclosure at the three-hour mark for adults.

Handle the age question. If your chatbot does not determine user age, decide between implementing an age gate or applying the one-hour disclosure interval to all users as a default. Document which approach you chose and why.

Build or audit minor protection filters. For any chatbot accessible to minors, review outputs for the six prohibited behaviors. Work with your AI vendor on filter settings or prompting strategies that prevent these outputs. This is not a one-time audit. Test it periodically.

Document your crisis response protocol. Write a description of what your chatbot detects, what response it generates (including the 988 referral), and how often the protocol is tested. A one-page document is enough. Review it when you update the underlying model or prompting system.

Review vendor SLAs. If you use a third-party chatbot platform such as Intercom, Drift, or Zendesk AI, ask your vendor directly whether they will implement SB 540 compliance features before July 2027. Most platforms will update their products, but you need written confirmation that their implementation covers your specific deployment, not just a general statement about compliance.

Set a July 2026 internal review date. Build in a year for implementation and one month for testing and audit before the July 1, 2027 effective date.

Georgia's SB 540 is part of a broader state-level movement to create behavioral guardrails around AI chatbots, with minor safety as the primary driver. The 13-month window is generous compared to the 6-month periods some states have used for AI disclosure requirements. The practical work of disclosure engineering, minor protection filters, and crisis response protocols takes longer to implement on third-party platforms than building from scratch.

For small teams using off-the-shelf chatbot platforms, the most important step is reaching your vendor now with a direct question: does your platform support SB 540 compliance features? If not, when will it?