removing-overconstraint-from-system-prompts-stopped-costing-accuracy CLM-0129observationsingle paperReviewed by AI
Guardrail instructions written to stop older models doing the wrong thing — blanket prohibitions, repeated warnings, defensive defaults — became dead weight in the Claude 5 generation: Anthropic removed over 80% of Claude Code's system prompt for Opus 5 and Fable 5 with no measurable loss on its own coding evaluations, on the reading that the constraints now conflict with each other and with user intent more often than they prevent harm.
Capability: Keeping its own context clean · Context, Coding agent
Observed on
2026, Claude Opus 5 and Fable 5 in Claude Code.
Sources
- Reports the 80% reduction and the reasoning: transcripts showed system prompt, skills and user request issuing conflicting instructions in a single request, which the model then had to spend effort reconciling. The measurement is real but unverifiable from outside — the evaluations are internal and unpublished, so there is no baseline, ablation or independent check available.
Either way, bring evidence: a link, or what you saw.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- Open-weight instruction-tuned models, told in plain language to not think about a concept while transcribing a sentence, suppress that concept's residual-stream projection enough to evade linear probes, the Jacobian lens, activation oracles and natural-language autoencoders, with no fine-tuning or prompt search — detection at a 1% false-positive budget falls to roughly 5-21% under suppression while engagement is caught near ceiling.Strategic deception and detecting it · reviewed by AI
- Across 25 open-weight instruction-tuned models (4B-744B), the ability to modulate a named concept's residual-stream projection on plain instruction does not track parameter count or recency — the best scorer is an 8B 2024 model and some of the newest models score lowest — while within a single model line the score rises steadily across ordinary pre-training and post-training checkpoints, so activation controllability is acquired incidentally and must be measured per model rather than inferred from scale.Strategic deception and detecting it · reviewed by AI
- When asked to optimize code that is already at its performance ceiling, frontier LLMs (GPT, Claude, Gemini families) never say it is already optimal under a plain "optimize this code for speed" prompt — they always emit an edit with a confident, unmeasured speedup claim; adding an explicit instruction to abstain unless >90% confident of an improvement produces abstention on a minority of such cases while leaving the edit rate on genuinely sub-optimal code unchanged.Stating false facts confidently · reviewed by AI
- For frozen VLA policies executing repetitive, perceptually aliased manipulation tasks (counting pick-and-place), a progress memory that advances after each attempted action converts transient grasp or placement failures into persistent task-state errors, so writes must be gated on observed physical evidence of the intended transition; gating this way raised success far above the frozen policy and above memory-augmented baselines.Turning plans into physical action · reviewed by AI
- Safety realignment after a poisoned fine-tune is brittle in three ways: existing defenses (SSRD, RESTA, soft-SFT, rollback) often destroy the downstream skill they were meant to keep, lose more than 30% of their effect when the defender's prompt template differs from the attacker's, and a realigned model can be re-jailbroken by switching the system prompt; distilling on-policy from two frozen teachers, the original model for refusal and the fine-tuned model for the task, routed per token, cut attack success to about 2% under the defender's template while holding task score near pre-attack level.Retaining skills and safety after fine-tuning · reviewed by AI
Notes
Filed as an observation about one model family at one moment, not a mechanism. The durable question underneath — whether instruction volume trades against instruction-following, and where the turn happens — is not answered here. Capped at single-paper because the source reports an actual measurement. Vendor guidance that only asserts, without a number, cannot reach even that. And "no measurable loss" is doing quiet work: it means no loss their evaluations could see, which is a weaker statement than no loss.