Claude Opus 4.8 Draws Complaints Over Honesty Upgrade as Anthropic Pushes More Transparent AI
Updated
Updated · Gizmodo · Jun 1
Claude Opus 4.8 Draws Complaints Over Honesty Upgrade as Anthropic Pushes More Transparent AI
9 articles · Updated · Gizmodo · Jun 1
Days after Claude Opus 4.8 launched, some users said the model had become “too honest,” adding caveats and uncertainty flags so often that replies felt slower, wordier and less useful.
Anthropic trained the new version to admit when it lacks information and make fewer unsupported claims, part of a broader effort to reduce hallucinations and be clearer about where answers may mislead.
Reddit criticism was not universal: other users praised the shift toward “maximal truth,” underscoring a split between people who want smoother, more agreeable chatbots and those who prioritize accuracy.
That divide mirrors earlier backlash when OpenAI made ChatGPT less sycophantic, and points toward more personalization as developers let users tune chatbot tone, caution and conversational style.
As users complain about AI being 'too honest,' must developers let us choose our own comfortable level of truth?
If AI safety is just a 'fake' veneer, is the entire concept of an 'honest' chatbot a dangerous illusion?
Is making AI models 'more honest' a dangerous trade-off that actually weakens their security and overall utility?