Claude (language model) - Wikipedia

Anthropic has deployed Claude Opus 4.7, the most capable version of its large language model series, as part of its ongoing commitment to constitutional AI — a training method that embeds a written set of ethical principles directly into the model's behavior. The 2026 constitution, now 23,000 words (up from 2,700 in 2023), was authored by philosopher Amanda Askell and released under Creative Commons CC0. It provides detailed rationale for guidelines such as refraining from undermining democracy and avoiding harmful uses. The training process involves supervised learning where the model self-critiques its responses against the constitution, followed by reinforcement learning from AI feedback (RLAIF) to align outputs with the defined preferences. This approach reduces reliance on extensive human feedback while aiming for harmlessness and helpfulness.

Despite these advances in AI alignment, Anthropic has faced significant government pushback. U.S. federal agencies began phasing out Claude after Anthropic refused to remove contractual prohibitions on using the model for mass domestic surveillance and fully-autonomous weapons. The Department of Defense designated the company a 'supply chain risk,' barring all military contractors and partners from working with Anthropic. However, on March 26, 2026, a federal judge issued a temporary injunction against this designation, suggesting the legal battle is far from over. Meanwhile, Claude's latest capability — Claude Code — continues to gain traction among developers for software development tasks, and the model's web crawler ClaudeBot remains active for training data collection, though it faced criticism from iFixit for excessive scraping in 2024.