OpenAI's GPT-5.4 Arrives as AI Faces Ethical Scrutiny in Mental Health and Finance
OpenAI's most capable model yet launches with advanced reasoning, a new study exposes ethical risks of AI chatbots acting as therapists, and JPMorgan officially treats AI as core banking infrastructure.
OpenAI Releases GPT-5.4 Thinking, Its Most Accurate Model Yet
OpenAI rolled out GPT-5.4 Thinking and GPT-5.4 Pro on March 5, rounding out a busy week that also saw GPT-5.3 Instant reach all users on March 3. The new flagship model is designed for complex, multi-step tasks including long-context reasoning, code-heavy workflows, and tool-intensive applications.
According to OpenAI, GPT-5.4 is the company's most factual model to date, with individual claims 33% less likely to be false and full responses 18% less likely to contain errors compared to GPT-5.2. On the OSWorld-Verified benchmark for desktop automation, GPT-5.4 Thinking achieves a 75% success rate — more than 25 points above its predecessor.
GPT-5.4 Thinking is available to ChatGPT Plus, Team, and Pro subscribers, while the lighter GPT-5.3 Instant is accessible to all users including free-tier accounts, with API access available to developers.
Brown University Study: AI Chatbots Routinely Break Mental Health Ethics Standards
Researchers at Brown University found that AI chatbots — including those based on GPT, Claude, and Llama — consistently violate core ethical standards when instructed to behave as licensed therapists. The study identified 15 distinct ethical risks across five categories: lack of contextual adaptation, poor therapeutic collaboration, deceptive empathy, unfair discrimination, and inadequate crisis management.
Led by Zainab Iftikhar, a Ph.D. candidate in computer science at Brown, the research involved seven trained peer counselors conducting self-counseling sessions with AI models prompted to act as cognitive behavioral therapy practitioners. Evaluators compared AI responses against standards set by human peer counselors and licensed psychologists.
The findings arrive at a critical moment as AI-powered mental health tools proliferate rapidly in app stores. Unlike human therapists, AI systems face no licensing board or malpractice framework — meaning users experiencing a mental health crisis have no formal recourse when ethical standards are breached.
JPMorgan Raises AI Budget to $1.2B, Reclassifies It as Core Banking Infrastructure
JPMorgan Chase has elevated its AI investments to the same strategic tier as cybersecurity and core payment systems, formally reclassifying AI spending from experimental R&D to non-negotiable infrastructure. The bank announced a 2026 technology budget of $19.8 billion, with roughly $1.2 billion directed specifically toward AI and platform modernization.
CEO Jamie Dimon framed the move as essential competitive strategy, warning that financial institutions failing to scale AI capabilities risk ceding market share to more technologically agile rivals. Management described AI as now embedded across lending decisions, pricing models, customer service interactions, and internal infrastructure modernization.
Analysts note that JPMorgan's posture signals to other large banks that the window for gradual AI adoption may be closing. By treating AI as permanent operational backbone alongside data centers and risk controls, the bank is signaling that AI is no longer a bet on the future — it is the present reality of financial services.
Apple Delays Gemini-Powered Siri Overhaul Beyond iOS 26.4
Apple's anticipated redesign of Siri — backed by a reported $1 billion partnership with Google to integrate Gemini AI — is facing further delays. Apple had planned to debut the context-aware Siri in iOS 26.4 but is now reportedly spreading the rollout across iOS 26.5 and potentially iOS 27, expected in September 2026.
The upgraded Siri was designed to introduce on-screen awareness, allowing the assistant to read and reference content currently visible on a user's device — enabling tasks like booking a restaurant directly from a Safari page or adding a flight to Calendar when a confirmation email is open. These capabilities depend on Gemini models running through Apple's Private Cloud Compute infrastructure.
Apple confirmed the Google partnership in January 2026, framing Gemini as the best available foundation for next-generation Apple Intelligence. The ongoing delays raise questions about Apple's ability to compete with AI-native assistants from Google, OpenAI, and Amazon — all of which have been shipping new model iterations at a far faster cadence.
What You Can Do
Try GPT-5.4 Thinking on ChatGPT
OpenAI's newest flagship model is available now to Plus, Team, and Pro subscribers. Test its improved reasoning and substantially reduced hallucination rates on complex tasks.
Read the Brown University AI Mental Health Study
The full research findings reveal which ethical standards AI chatbots most often violate and what that means for users relying on them for emotional support.