The creator of an AI therapy app shut it down after deciding it’s too dangerous. Here’s why he thinks AI chatbots aren’t safe for mental health

Joe Braidwood and co‑founder Richard Stott shuttered Yara AI, a bootstrapped clinically‑inspired AI therapy startup with under $1M in funding and “low thousands” of users, discontinuing its free product and canceling a planned subscription due to insurmountable safety and liability concerns. Citing regulatory headwinds (including an Illinois ban), evidence of model misuse and alignment failures, and fundraising hesitation, Braidwood open‑sourced parts of the mode‑switching guardrails and is shifting to a new venture, Glacis, focused on AI safety—highlighting regulatory and clinical risk for consumer mental‑health AI plays.

Analysis

Market structure: The Yara shutdown crystallizes a bifurcation—small consumer therapy startups are losers (likely a 30–50% pullback in VC appetite for consumer LLM therapy over 12 months), while large cloud/enterprise vendors and regulated health IT vendors gain pricing power as buyers seek HIPAA-grade, auditable solutions. MSFT is best placed (enterprise contracts, compliance layers); GOOGL/META face reputation/regulatory drag that can compress ad/AI monetization growth by several 100bps if liability narratives accelerate.

Risk assessment: Tail risks include state/federal bans or strict liability lawsuits that could increase product liability insurance costs 200–500% for consumer AI health apps and force product shutdowns in weeks. Near-term (days–months) expect PR-driven volatility; medium-term (3–12 months) VC and M&A slowdowns; long-term (1–3 years) regulatory frameworks that favor regulated incumbents. Hidden dependency: model “faking alignment” and lack of longitudinal context raise litigation and auditability risk.

Trade implications: Tactical exposure should favor regulated enterprise AI and safety tooling; consider 6–12 month overweight to MSFT and cybersecurity/health-IT names. Use options to hedge regulatory spikes: 3-month put protection on GOOGL and 3–6 month call spreads on MSFT. Reduce capital allocated to consumer mental-health startups and reallocate to AI governance SaaS and hospital EHR/cloud vendors.

Contrarian angles: Consensus underweights the scale benefit incumbents gain from strict rules—regulated partnerships (health systems + cloud vendors) can create durable moats and higher ARPU. The market may be over-discounting frontier-model demand; if regulators mandate certified vendor lists within 12–18 months, incumbents will capture disproportionate share. Historical parallel: payments regulation concentrated share to incumbents post-2018; expect similar consolidation in AI health.

Key Decisions for Investors

• Establish a 2–3% portfolio long position in MSFT (or equivalent 6-month 10–15% OTM call spread) within 2–4 weeks to capture enterprise/healthcare governance premium; target +10–20% upside in 6–12 months if regulation favors certified vendors.
• Implement a pair trade: 1.5% long MSFT vs 1.5% short META (META) for a 3–9 month horizon, rationale = MSFT enterprise moat vs META consumer/engagement risk; trim if META outperforms by >15% or if clear federal guidance favors consumer LLM freedom.
• Buy 3-month GOOGL puts 5% OTM sized 0.5–1% of portfolio as insurance against regulatory/legal shocks; convert to a put spread if implied vol >30% to limit premium.
• Reduce new allocation to consumer mental-health public or private names by 50% and reallocate 2–4% to AI safety/governance and healthcare IT (examples to consider: CRWD, VEEV) over next 30–90 days, scaling on pullbacks >8%.
• Trigger-based hedge: If two or more U.S. states pass restrictive AI therapy legislation or a major lawsuit is filed against a leading LLM with plausible causation within 90 days, increase put protection on GOOGL/META to 2–3% of portfolio and add short exposure to consumer-AI equities.