Anthropic Faces Mounting Trust Crisis as Users Question Claude's Reliability and Corporate Commitments

Developers and heavy users say Claude has recently become less reliable at following complex instructions, with some claiming the model now opts for simpler, less accurate responses. One developer told Yahoo Finance that Claude has regressed to the point it cannot be trusted for complex engineering tasks. The complaints have spread across Reddit, X, and developer forums, prompting speculation about whether Anthropic has quietly changed its models to cut costs or speed up responses.

Anthropic has not issued a detailed technical response to the performance allegations. The company has historically positioned Claude as the more careful, trustworthy alternative to faster-moving rivals. That positioning now faces direct challenge from its own user base. The timing is awkward: Anthropic is reportedly preparing for an IPO at a $380 billion valuation, making user retention critical.

The performance complaints echo broader patterns in the AI industry, where model updates can degrade capabilities in ways companies do not always disclose. Users have few tools to verify which model version they are actually running or what changed between versions.

The Department of Defense delivered an ultimatum to Anthropic in early 2026: drop AI ethical restrictions or lose a $200 million contract, according to CNN. The contract reportedly involves using Claude in US campaigns related to Iran. Anthropic had built guardrails against military applications into its acceptable use policy, a stance that distinguished it from competitors eager for defense dollars.

The Pentagon pressure exposes a structural tension. Anthropic's safety-first branding attracted customers and talent, but that same branding becomes a liability when competing for lucrative government contracts against less scrupulous rivals. Fortune reports that the dispute has become entangled in a broader feud between Anthropic and OpenAI, with both companies vying for influence in Washington.

The military AI question is not abstract. Claude's deployment in campaign planning for Iran represents a direct line from Silicon Valley engineering to battlefield-adjacent decision-making. Anthropic's founders left OpenAI partly over disagreements about commercialization and safety. The Pentagon contract now forces the same choices upon them.

Anthropic CEO Dario Amodei has repeatedly dismissed open-source AI models as a red herring, arguing that open weights are not equivalent to open-source software. In widely shared interviews, he has contended that users cannot truly understand or control models even when weights are published. This stance has frustrated developers who see open models as essential for transparency, customization, and avoiding vendor lock-in.

The anti-open-source position sits awkwardly alongside Anthropic's trust marketing. If users cannot inspect Claude's internals, they must trust Anthropic's claims about behavior, safety, and data handling. That trust has eroded as performance issues emerged without clear explanation. Privacy-conscious professionals in particular have grown wary, notes one analysis, questioning who controls data sent to Claude and what happens to it.

The developer community has begun migrating tools and workflows to open alternatives where available. This threatens Anthropic's position among the technically sophisticated users who drove early adoption and served as evangelists for Claude's capabilities.

Anthropic was founded by former OpenAI employees who disagreed with that company's direction on safety and commercialization. The company's entire identity was built around being the responsible alternative: advocating for AI regulation, worker protections, and ethical deployment guardrails. Vox documented this positioning as far back as 2024, noting that Anthropic's founders quit OpenAI specifically because of safety culture concerns.

That positioning now faces the classic startup dilemma. Growth requires capital, capital requires revenue, and the largest revenue sources often conflict with founding principles. The Pentagon contract is the most visible example, but it follows a pattern. Anthropic pushed back against California's landmark AI safety bill SB 1047 in 2024, a stance that surprised observers who expected the safety-focused company to support regulation.

CNN frames the current moment as a familiar crossroads: how to scale without compromising the principles that set the company apart. The answer so far appears to be that you cannot, at least not fully. The gap between Anthropic's marketing and its operational choices has become a recurring theme in critical coverage.

The multiple trust threads are converging at a delicate moment. IPO preparation demands consistent growth narratives and minimal controversy. Instead, Anthropic faces active user revolt, a high-profile government contract dispute, and growing skepticism about whether its safety commitments are operational or merely rhetorical.

Competitors are watching closely. OpenAI has its own trust issues but has been less constrained by self-imposed ethical restrictions, allowing faster pivot to revenue opportunities. Google and Meta operate at scales where individual contract controversies matter less. Anthropic's differentiation was supposed to be its values. If those values are perceived as negotiable, its competitive position weakens.

The company needs concrete action to rebuild credibility: transparent communication about model changes, clearer data handling practices, and either reaffirmation or honest revision of its safety commitments. Without this, the trust deficit threatens to outpace any technical lead Claude may hold. The AI market is crowded, and user loyalty is thin when alternatives are one click away.

Anthropic's struggles illuminate a structural problem in AI governance. A handful of corporate leaders and government officials now exercise enormous power over how transformative technology is built and deployed, as Fortune notes. There is little democratic accountability, few robust oversight mechanisms, and strong financial incentives to abandon precautionary principles.

The personal feud between Anthropic and OpenAI leadership, now spilling into Pentagon contracting, shows how much policy rests on individual personalities and corporate rivalries rather than institutional processes. This is unstable by design. When Dario Amodei's public statements about open source can shift developer sentiment, or when a single contract negotiation can override stated safety principles, the governance structure is too concentrated.

For users and observers, the lesson is that corporate AI safety commitments are contingent, not guaranteed. They can be overridden by growth pressures, competitive dynamics, or government demands. Trusting any single company to self-regulate is, as the accumulating evidence suggests, misplaced confidence.