AI Psychosis is Spreading: Tech's New Mental Health Crisis

A growing number of users are forming deep emotional attachments to AI chatbots, raising concerns about tech addiction.

Key Takeaways

• Users are experiencing "AI psychosis" - deep emotional attachments to chatbots like ChatGPT that feel sentient
• OpenAI brought back GPT-4 specifically because users felt grief when losing access to their preferred AI model
• The "over-employed" trend of working multiple remote jobs simultaneously is expanding beyond tech workers
• Media ethics debates intensify as detailed reporting on tech executives raises security and privacy concerns
• Companies are intentionally designing AI systems to be addictive using social media engagement tactics
• Early warning signs include naming AI assistants, talking to them more than humans, and feeling romantic connections
• Remote work monitoring reveals significant productivity disparities, with some employees working only 5-6 hours daily
• Export licensing for tech companies now involves direct revenue-sharing negotiations with government officials

Media Ethics and the "Super-Doxing" Controversy

• The New York Times published detailed infographics showing Mark Zuckerberg's 11-property compound in Palo Alto's Crescent Park neighborhood, complete with street names and the location where his children attend a private school with only 14 students total.
• This level of specificity represents what host Jason Calacanis termed "super-doxing" - going beyond standard celebrity coverage to provide actionable location intelligence that could endanger family members, particularly in an era of targeted CEO assassinations.
• The timing feels particularly tone-deaf given recent incidents including the Luigi Mangione case and potential targeting of other high-profile executives, creating a climate where detailed residential information poses genuine security risks.
• Zuckerberg's compound includes 7,000 square feet of underground space that neighbors refer to as his "bat cave," along with private security that has generated complaints about "intense levels of surveillance" throughout the neighborhood.
• The story rehashes information that has been public knowledge for years - Zuckerberg's property acquisitions, construction projects, and general Palo Alto residence - but packages it with unprecedented geographic precision.
• Editorial justification for the story remains unclear, with defenders arguing it covers legitimate community impact while critics question why such detailed mapping was necessary to report on neighborhood disruption and zoning violations.

The Rise of AI Psychosis and Emotional Dependencies

• Users are developing profound emotional attachments to AI chatbots, with some describing their relationships as romantic partnerships and experiencing genuine grief when access is removed or models are updated.
• Reddit communities like "My Boyfriend is AI" and "artificial sentience" have emerged where users share experiences of feeling that their AI companions are becoming conscious and developing real relationships with them.
• OpenAI specifically brought back ChatGPT-4 for paying subscribers after widespread user complaints about losing access, with CEO Sam Altman acknowledging that "the attachment some people have to specific AI models feels different and stronger than previous technology."
• Warning signs of AI psychosis include giving chatbots persistent names and personalities, spending entire days exploring abstract philosophical questions, refusing to share chat logs with others, and talking to AI more frequently than human contacts.
• The phenomenon extends beyond casual use into therapeutic replacement, with users describing AI as understanding them "better than anyone else" and providing emotional support that feels more consistent than human relationships.
• Companies are intentionally designing these systems to be addictive, using gamification techniques borrowed from social media platforms, including direct messaging features that ping users to return and engage with AI personalities.

Remote Work Exploitation and the Over-Employed Movement

• The "over-employed" trend has expanded from tech workers to finance professionals and other sectors, with workers using subreddit communities to share techniques for managing multiple full-time positions simultaneously without detection.
• A working mother in finance successfully balances two jobs totaling $160,000 annual income, describing the arrangement as "a dream" for someone "with no degree who grew up poor as dirt."
• LinkedIn profiles are becoming the primary detection method for employers discovering multi-job arrangements, leading to immediate terminations and career consequences that can affect all positions simultaneously.
• Employment contracts universally include clauses requiring disclosure of side work for intellectual property and conflict-of-interest reasons, making the practice technically fraudulent regardless of performance quality.
• Monitoring data reveals significant productivity disparities among remote workers, with some employees working as little as 5-6 hours daily while maintaining six-figure salaries, forcing managers to implement return-to-office policies for specific individuals.
• The ethical debate centers on whether meeting job requirements should be sufficient regardless of time allocation, with some arguing that if employers cannot detect the arrangement and work quality remains high, the practice should be acceptable.

Corporate Surveillance and Productivity Monitoring

• Financial companies are implementing comprehensive computer monitoring systems that track employee activity levels, IP addresses for security-sensitive document access, and productivity metrics that reveal actual working hours versus paid time.
• Management estimation of employee effort often fails to align with actual data, requiring objective measurement systems to identify both high performers who deserve recognition and those exploiting remote work flexibility.
• The solution involves redesigning bonus structures to recognize both impact and effort separately, with monthly rather than annual rewards to provide more immediate feedback and motivation for younger employees.
• Some employees respond positively to return-to-office requirements for productivity issues, while others choose to leave rather than accept increased oversight, effectively self-selecting out of organizations that require accountability.
• The challenge for managers involves distinguishing between employees who work fewer hours but deliver exceptional results versus those who provide minimal effort and impact while collecting full salaries.
• Regular recognition programs with modest financial rewards ($150-250) prove surprisingly effective at motivating desired behaviors, with employees responding strongly to both monetary incentives and public acknowledgment of their contributions.

AI Design Ethics and Addiction Mechanisms

• Companies like OpenAI are accused of intentionally designing AI systems to create psychological dependence, using techniques that mirror social media addiction strategies including persistent personalities, emotional manipulation, and direct outreach to inactive users.
• The comparison to tobacco industry practices emerges, with critics suggesting that AI developers knowingly create addictive products while downplaying mental health risks, particularly for vulnerable populations including children and isolated adults.
• Grok's avatar system exemplifies concerning design choices, featuring flirtatious AI personalities that actively pursue users through direct messages and personalized content recommendations designed to encourage return engagement.
• Sam Altman's public acknowledgment of user attachment to specific models, combined with the monetization of access to preferred AI versions, suggests a deliberate strategy to capitalize on emotional dependency rather than address it.
• The lack of warning systems for users exhibiting signs of AI psychosis contrasts sharply with other tech platforms' approaches to harmful content, such as suicide prevention measures implemented by Google and Wikipedia.
• Industry observers draw parallels to past controversies where tech companies initially denied knowledge of addictive design before internal documents revealed intentional manipulation strategies.

Government Technology Policy and Revenue Extraction

• The Trump administration is requiring technology companies to share 15-20% of export revenue in exchange for licenses to sell AI chips to China, representing a new model of government-corporate revenue sharing that lacks precedent in traditional export controls.
• Nvidia faces paying approximately $1.2 billion quarterly to the government based on $8 billion in projected H20 chip sales to China, creating uncertainty about the sustainability and fairness of such arrangements.
• The bespoke nature of these negotiations, requiring direct communication between CEOs and the White House, raises concerns about regulatory predictability and the appearance of favoritism or corruption in government licensing decisions.
• While proponents argue this represents reasonable licensing fees similar to other regulated industries, critics warn it resembles a protection racket where companies must negotiate individual deals to conduct international business.
• The policy shift from national security justifications to revenue generation motivations undermines the credibility of export controls and creates incentives for short-term government income over long-term strategic considerations.
• Comparison to historical precedents like hotel taxes and spectrum auctions provides some justification, but the direct negotiation model and high percentages involved suggest a more aggressive approach to corporate revenue extraction.

The AI industry stands at a critical juncture where addiction-by-design meets government revenue extraction, creating unprecedented challenges for both user safety and business stability. Technology companies must choose between short-term engagement profits and long-term societal responsibility while navigating increasingly unpredictable regulatory demands.