Anthropic platform trust gaps
- Anthropic updated Claude Opus 4.7 docs and pricing, but its Labs release reportedly lacks audit logs and usage tracking. - Cybersecurity reports and coverage flagged Claude producing vulnerable code and confusion over product plan inclusions. - Those gaps underline that model quality alone won't satisfy enterprise buyers; auditability, stable plans, and secure defaults are also required (platform.claude.com) (forbes.com) (theregister.com).
Anthropic spent the past week pitching Claude Opus 4.7 as a stronger coding model while customers ran into missing controls, shifting plan language, and new security complaints. (anthropic.com) (support.claude.com) (forbes.com) Anthropic announced Claude Opus 4.7 on April 16, 2026, said pricing stayed at $5 per million input tokens and $25 per million output tokens, and added features including a 1 million-token context window, 128,000 max output tokens, high-resolution image support, and a new “xhigh” effort setting. (anthropic.com) (platform.claude.com) A day later, Anthropic launched Claude Design in research preview through Anthropic Labs for Pro, Max, Team, and Enterprise subscribers. Its help center says the product “doesn’t support audit logs or usage tracking yet,” even as Anthropic separately offers audit logs for Enterprise and analytics for Team, Enterprise, and API Console users in other Claude products. (anthropic.com) (support.claude.com 1) (support.claude.com 2) (support.claude.com 3) (support.claude.com 4) The same week, Anthropic’s subscription pages and product pages pointed in different directions on what paid plans included. The Register reported on April 22 that Claude Code disappeared from some Pro-plan pricing language, while Anthropic growth head Amol Avasare said the company was running a test on about 2 percent of new prosumer signups and that existing Pro and Max subscribers were not affected. (theregister.com) Anthropic has also been reworking business pricing. The Register reported on April 16 that the company had been renewing enterprise customers onto usage-based plans since November 2025, after introducing a $20-per-employee monthly enterprise seat in February 2026 that covered chat, Code, and Cowork on a metered basis. (theregister.com) Those plan changes landed as Anthropic marketed Opus 4.7 as better for “advanced software engineering” and said users could hand off harder coding work with more confidence. In the same announcement, Anthropic said Opus 4.7 was the first model released with new safeguards to detect and block prohibited or high-risk cybersecurity requests, and it opened a Cyber Verification Program for legitimate security researchers. (anthropic.com) Outside testing has been less flattering. Forbes reported on April 22 that TrustedSec CEO Dave Kennedy said Claude’s code quality had fallen 47.3 percent from its earlier level and that Opus 4.7 was only “marginally better,” while Veracode told Forbes that Opus 4.7 produced vulnerable code in 52 percent of 80 coding tasks in its testing. (forbes.com) Anthropic has publicly acknowledged the security tension around stronger coding systems. In its April 16 launch post, the company said Opus 4.7 is less capable than its unreleased Mythos Preview model and said it had experimented during training with reducing some cyber capabilities before release. (anthropic.com) The result is a split screen: Anthropic’s latest model arrived with higher-end specs and broader product rollouts, while the surrounding platform still shows gaps in logging, analytics, and plan clarity. For buyers comparing AI vendors in April 2026, those operational details are now showing up alongside benchmark scores and model demos. (platform.claude.com) (support.claude.com) (theregister.com)
