Just days after releasing GPT-5.3 Instant, OpenAI has unveiled GPT-5.4, a frontier model that combines reasoning, coding, and computer control capabilities in a single package. But this technical achievement arrives amidst a brewing storm over the company’s decision to partner with the U.S. Department of Defense, a move that has sparked user backlash and intensified competition with rival Anthropic.
The Technical Leap Forward
GPT-5.4 represents more than just another incremental update. For the first time, OpenAI has integrated native computer-use capabilities directly into a general model, allowing AI agents to navigate desktop environments, control mouse and keyboard inputs, and execute complex workflows across multiple applications without specialized add-ons. On the OSWorld-Verified benchmark for agentic desktop control, GPT-5.4 achieves 75%, surpassing both the human reference score of 72.4% and Anthropic’s Opus 4.6, which previously set the standard at 72.7%.
The improvements extend across multiple domains. On GDPval, which measures agent performance across 44 professional fields, GPT-5.4 achieves an 83% win rate against industry experts, significantly outperforming GPT-5.2’s 70.9%. For investment banking modeling tasks, the new model scores 87.3% compared to 68.4% for its predecessor. OpenAI also claims substantial reductions in hallucination rates, with individual statements being 33% less likely to contain errors and complete answers having 18% fewer mistakes.
The Military Contract Controversy
While OpenAI celebrates technical milestones, the company faces mounting criticism over its recent Pentagon contract. According to TechCrunch reporting, OpenAI won a Department of Defense deal that Anthropic had just walked away from due to ethical concerns about mass surveillance and automated killing. The backlash has been immediate and measurable: ChatGPT mobile app uninstalls surged 295% day-over-day on February 28, 2026, while competitor Anthropic’s Claude app saw downloads jump 51% on the same day.
OpenAI CEO Sam Altman attempted to address concerns in a public Q&A on X, stating, “I very deeply believe in the democratic process, and that our elected leaders have the power, and that we all have to uphold the constitution.” However, the Financial Times reports that OpenAI amended its Pentagon contract just days after signing it, adding terms to prohibit domestic surveillance of U.S. persons and exclude intelligence services like the NSA. Altman admitted the rushed process “looked opportunistic and sloppy.”
The Anthropic Counterpoint
Anthropic’s refusal to grant the DoD unrestricted access to its AI technology without safeguards has positioned the company as an ethical alternative. In a memo to staff reported by The Information, Anthropic CEO Dario Amodei accused OpenAI’s messaging around its military deal as “straight up lies” and “safety theater.” Amodei claims, “The main reason [OpenAI] accepted [the DoD’s deal] and we did not is that they cared about placating employees, and we actually cared about preventing abuses.”
The contrasting approaches have created a clear market response. While GPT-5.4 offers superior technical specifications and competitive pricing at $2.50 per million input tokens versus Opus 4.6’s $5, consumer sentiment appears to be shifting toward companies prioritizing ethical boundaries. Claude reached No. 1 on the U.S. App Store on February 28, 2026, while ChatGPT saw 1-star reviews surge 775% on the same day.
Enterprise Implications and Security Concerns
Beyond the military controversy, GPT-5.4’s enhanced capabilities raise important questions about enterprise security. According to ZDNET analysis, enterprise AI agents could become “the ultimate insider threat.” With machine identities now outnumbering human identities 82 to 1 in enterprises, and 72% of employees regularly using AI tools on the job, security vulnerabilities become magnified. Gartner estimates that more than 40% of enterprise apps will use AI agents in 2026, up from less than 5% in 2025, yet only 6% of organizations have an advanced AI security strategy.
GPT-5.4’s new “Tool Search” feature, which dynamically retrieves tool definitions rather than loading them all at once, reduces token consumption by 47% in tests with 36 MCP servers and 250 tasks. While this represents significant cost savings for tool-intensive applications, it also introduces new attack vectors that enterprises must consider.
The Competitive Landscape Reshaped
The simultaneous technical advancement and ethical controversy highlight a fundamental tension in today’s AI industry. OpenAI’s GPT-5.4 demonstrates clear technical superiority across multiple benchmarks, including abstract pattern recognition where the Pro variant achieves 83.3% on ARC-AGI-2, compared to Google’s Gemini 3.1 Pro at 77.1% and Opus 4.6 at 68.8%. Yet this technical edge comes with reputational costs that may influence enterprise adoption decisions.
As Wendi Whitmore, Palo Alto Networks chief security intel officer, notes, “the AI agent itself [is] becoming the new insider threat.” This warning takes on new urgency with models like GPT-5.4 that can autonomously control computer systems. The question for businesses becomes not just which model performs better, but which company they trust with increasingly sensitive operations.
The rapid-fire release cycle – GPT-5.4 arriving just two days after GPT-5.3 Instant – demonstrates the intense competition driving innovation. But as the military contract controversy shows, technical prowess alone may not determine market leadership in an industry where trust and ethics are becoming increasingly important differentiators.