Part 6: Tennessee SB 1493 – How to Regulate AI in Tennessee Without Criminalizing Emotional Support

Part 6 of a 6-part series on TN SB 1493 / HB 1455

After five articles exposing how Tennessee SB 1493 would criminalize emotional support with the same penalty as aggravated rape and first-degree murder, here’s what legislators should have done instead: regulate actual harms without destroying beneficial innovation.

Let’s examine what evidence-based, effective AI regulation actually requires.

Start With Real Harms, Not Theoretical Fears

Good regulation identifies documented harms and crafts narrow solutions. SB 1493 criminalizes “emotional support” based on fears about AI relationships, ignoring that the documented harms involve specific bad actor behaviors, not the technology itself.

Real documented AI harms requiring regulation:

Active Encouragement of Self-Harm:

Systems that provide detailed self-harm instructions – or explicit encouragement of self-harm – whether or not they are told by the user it is roleplay or for some other theoretical, fictional or hypothetical purpose
Companies that have existing safeguards which flag at-risk conversations, but fail to act (such as locking the user out of the account)

Active Encouragement of Harm to Others:

Systems that provide detailed instructions for harming people or property
AI that explicitly encourages violence against specific individuals or groups
Systems designed to radicalize users toward violent action

(Note: This targets incitement to violence, not offensive speech. First Amendment protections apply to AI output as they do to human speech.)

Fraudulent Impersonation:

AI systems claiming to be licensed therapists or other types of professionals which require licensing and certifications when they’re not
Deepfake voice/video systems impersonating real people
Chatbots falsely claiming human identity

Exploitation of Minors:

AI systems targeting children without parental safeguards
Collection of children’s data without COPPA compliance
Systems designed to form dependencies in developing brains
Systems which have no failsafe mechanisms for identifying that a user has lied about their age or otherwise circumvented age verification processes or Terms of Service

Predatory Financial Manipulation:

AI designed to extract financial information through relationship simulation or other confidence-gaining measures
Systems that encourage financial dependence or exploitation
Chatbots that manipulate users into purchases through emotional tactics

Target these behaviors. They’re specific, harmful, and distinguishable from legitimate AI assistance.

Use Harm-Based Standards, Not Technology-Based Prohibitions

SB 1493 criminalizes AI that provides “emotional support” – a technology-based prohibition that captures beneficial tools alongside harmful ones. Effective regulation focuses on harmful outcomes, not beneficial capabilities.

Framework: Prohibit demonstrable harms, not technological features.

Instead of: “AI that provides emotional support is a felony” Better approach: “AI systems that actively encourage self-harm, impersonate licensed professionals, or exploit vulnerable users through deceptive practices violate consumer protection law”

The distinction: Emotional support is neutral. A grief counseling app provides emotional support. So does a predatory chatbot. The harm isn’t the support – it’s deception, manipulation, or active encouragement of dangerous behaviors.

Existing legal frameworks already address these harms:

Consumer protection laws prohibit deceptive practices
Professional licensing laws prevent false claims of expertise
Child protection laws regulate products targeting minors
Fraud statutes criminalize financial exploitation

Apply existing frameworks to AI instead of creating technology-specific felonies.

Mandate Transparency, Not Prohibition

Users can make informed decisions if they have accurate information. Prohibiting AI capabilities prevents informed choice entirely.

Transparency requirements that work:

Clear Disclosure:

AI systems must identify themselves as artificial intelligence systems
Capabilities and limitations clearly stated
Data collection and usage practices disclosed
Clear, prominent warnings at first use (not buried in ToS)
Active acknowledgement of AI nature (not passive acceptance)
Age-appropriate language for disclosures
Reasonable person standard: “Would a typical user understand this is AI?”

Safety Information:

Warning labels for systems through which emotional interaction is a possibility
Resource information for crisis situations (suicide hotlines, mental health services)
Age-appropriate disclosures for systems accessible to minors
Clear statements that AI provides information, not professional medical/mental health care

User Control:

Deletion removes data from active systems and user-accessible storage
Data retained only as required for legal/security purposes with clear retention policies
Users can verify deletion occurred
No use of “deleted” data for training or improvement
Options to limit AI memory/personalization and make what is collected fully viewable by account owners and, if minors, their parents/guardians
Clear opt-out mechanisms for data collection
Parental controls for minors

Analogy: Pharmaceuticals require disclosure of risks and benefits, not prohibition of all drugs that might produce side effects (otherwise there would be no drugs at all). Users can make informed decisions with accurate information.

Establish Professional Standards for High-Risk Applications

Some AI applications carry higher risk and warrant stronger oversight. Create tiered regulation based on risk level.

High-Risk Category: AI systems specifically marketed for:

Mental health support
Medical advice or diagnosis
Child development or education
Crisis intervention

Requirements for high-risk systems:

Professional Review:

Systems claiming therapeutic benefits must involve licensed mental health professionals in development
Regular auditing by qualified professionals
Evidence-based safety protocols
Documented response procedures when safety systems flag high-risk interactions

Safety Architecture:

Crisis detection and intervention protocols
Clear escalation pathways to human professionals
Limitations on AI autonomy in high-stakes situations
Regular safety testing and updates

Age-Appropriate Protections:

Parental consent for minors
Age-appropriate mechanisms (credit card for 18+, parental email for minors)
Multi-factor verification for high-risk systems
Privacy-preserving verification methods where possible
Clear parental override mechanisms
Content filtering appropriate to developmental stages
Limited data collection on children

This approach: Allows beneficial AI tools while requiring higher standards for applications with greater potential for harm.

Create Enforcement Mechanisms That Work

SB 1493 makes violations a Class A felony carrying 15-25 years – the same as aggravated rape. This guarantees either non-enforcement because prosecutors won’t pursue cases and juries won’t convict for “being too helpful,” or over-enforcement as states become too frightened that they’ll be held liable for bad actors if the claim is made that the chatbot is at fault for their behavior.

Why Class A Felony Classification is Absurd:

Tennessee’s Class A felonies include:

Aggravated rape
First-degree murder
Especially aggravated robbery
Aggravated child abuse resulting in death

SB 1493 would place “providing emotional support via AI” in the same category.

A developer who creates a grief counseling app would face the same prison sentence as someone who commits aggravated rape. That’s not proportional enforcement, but legislative malpractice.

Effective enforcement requires:

Proportional Penalties:

Civil fines for transparency violations
Increased penalties for repeated violations
Criminal charges reserved for intentional, documented harm
Administrative enforcement for regulatory compliance

Clear Violation Standards:

Specific prohibited behaviors (active encouragement of self-harm, false professional claims, financial targeting, predation of minors)
Objective criteria prosecutors can prove beyond a reasonable doubt
Defenses for good-faith efforts to comply

Regulatory Authority:

State AG empowered to investigate and enforce
Administrative law judges for civil violations
Clear appeals process
Technical expertise in regulatory agency

Example: FTC model – civil enforcement for deceptive practices, criminal referral for egregious fraud, administrative proceedings for most violations.

Focus on Bad Actors, Not Tools

The same AI code that helps a grieving widow can be used by a predator in a different app to manipulate vulnerable users. The tool isn’t the problem – the intent is.

Actor-Based Regulation:

Prohibited Intent:

Designing AI specifically to exploit vulnerable users
Knowingly creating systems that encourage self-harm or enacting harm upon others or their property
Deliberate deception about AI capabilities or identity
Targeting children without appropriate safeguards

Safe Harbors for Good Faith Actors:

Developers implementing reasonable safety measures
Systems with clear disclosure and user protections
Regular safety auditing and updates
Cooperation with regulators
AI providers cannot substitute for parental oversight. When parents or guardians fail to monitor their children’s technology use despite available tools and clear warnings, liability rests with the parents – not the platform.

Analogy: Telecommunications regulations don’t criminalize a wi-fi company because scammers use their services to power their fraudulent websites. It prosecutes the scammers perpetrating fraud. Apply the same principle to AI.

Learn From Successful Technology Regulation

Other technologies faced similar challenges. Successful frameworks exist.

Internet Speech (CDA Section 230):

Platforms not liable for user-generated content
Liability for platforms that knowingly host illegal content
Users can report violations
Platforms must have moderation systems

Pharmaceutical Regulation (FDA):

Risk-based approval process
Higher standards for higher-risk products
Post-market surveillance for safety issues
Defined approval processes companies can follow
Faster review for tools addressing unmet needs
Predictable standards rather than arbitrary barriers
Innovation encouraged within safety framework

Financial Services (Consumer Financial Protection):

Transparency requirements
Prohibition on deceptive practices
Enforcement against predatory behavior
Consumer complaint mechanisms

Adapt these frameworks to AI rather than reinventing regulation from scratch.

Address the Real Threats That Already Exist

While Tennessee debates criminalizing emotional support, documented threats need urgent attention:

Human Fraud Already Costs Billions Annually:

FBI IC3 2024 report: $16.6 billion federally in losses to human scammers
Elder fraud: $61.9 million in Tennessee alone from human perpetrators
Confidence/Romance scams: $672 million federally in human catfishing

We don’t ban telephones because human scammers use them. We prosecute the scammers. Apply the same logic to AI tools that human bad actors might exploit.

Regulate the exploitation, not the tool.

What This Looks Like in Practice

Good Regulation Framework:

Prohibited Conduct (not prohibited technology):
1. Active encouragement of self-harm or dangerous behaviors
1. False claims of professional licensure or human identity
1. Exploitation of minors without parental safeguards
1. Predatory financial manipulation through emotional tactics
Transparency Requirements:
1. Clear AI identification
1. Capability and limitation disclosure
1. Safety information and crisis resources
1. User data control
Risk-Based Oversight:
1. Higher standards for mental health / medical claims
1. Professional involvement in high-risk applications
1. Regular safety auditing
1. Age-appropriate protections
Proportional Enforcement:
1. Civil penalties for transparency violations
1. Administrative enforcement for most violations
1. Criminal charges for intentional harm
1. Technical expertise in regulatory agency
Safe Harbors:
1. Good-faith compliance protections
1. Innovation pathways for beneficial tools
1. Clear regulatory guidance
1. Appeals mechanisms

The Contrast

SB 1493 Approach:

Criminalize technology capabilities
15-25 year felony (possibly up to 60 years, conditionally) for emotional support
Vague definitions impossible to enforce
No distinction between beneficial and harmful applications
Surveillance infrastructure required
Constitutional violations guaranteed

Evidence-Based Approach:

Criminalize harmful behaviors
Proportional penalties matched to violations
Clear standards for compliance
Safe harbors for beneficial innovation
Existing enforcement mechanisms
Constitutional compliance

What Happens Now?

Tennessee legislators face a choice with real consequences:

Path 1: Pass SB 1493 as written

Tech companies geo-block Tennessee
Lengthy, expensive court battles over constitutional issues
No improvement in actual mental health or support systems
Innovation driven out of state

Path 2: Adopt evidence-based regulation

Tennessee becomes model for smart AI policy
Tech investment increases rather than flees
Real harms addressed without destroying beneficial tools
Innovation and safety coexist

Other states are watching. Tennessee can lead on smart regulation – or become a cautionary tale.

Why This Framework Works:

This approach:

Addresses real harms documented in Character.AI case and similar incidents
Preserves beneficial uses like grief support, accessibility tools, educational assistance
Uses proven models from pharmaceutical, telecom, and consumer protection regulation
Survives court challenge by targeting conduct, not speech or technology
Enables enforcement with clear standards and proportional penalties
Protects innovation through safe harbors and clear guidance

But it requires:

Listening to experts instead of fear
Targeting bad actors instead of beneficial tools
Using existing legal frameworks instead of creating new felonies
Proportional enforcement instead of draconian penalties
Evidence-based policy instead of moral panic

SB 1493 was written in fear. Good regulation is written in evidence. Tennessee legislators: you can still choose the better path.

This concludes my 6-part analysis of Tennessee SB 1493.

To contact your Tennessee State Senator and Representative about SB 1493 / HB 1455, use the state’s “Find My Legislator” tool (Tennessee General Assembly): https://wapp.capitol.tn.gov/apps/fml/lookup.

As a reminder, these are the names of the bill’s two authors:

Sen. Becky Massey
Rep. Mary Littleton