If every judge follows the rules and they all agree, can the group still sign off on something wrong?
Can a quorum of protocol-compliant validators certify a semantically invalid state?
This explores whether a group of validators that all follow the rules (they authenticate, sign the expected messages and reach consensus) can still approve a state that is actually wrong.
This explores whether a group of validators that all follow the rules (they authenticate, sign the expected messages and reach consensus) can still approve a state that is actually wrong. The corpus says yes. One note frames it as the 'honest quorum problem': validators can do everything the protocol asks and still endorse a semantically invalid transition. Can a quorum of honest validators certify an invalid transition? sums it up as 'agreement proves the protocol ran, not that the answer is right.'
The reason is that the guarantees come in two different kinds. According to Can validator consensus guarantee both agreement and semantic correctness?, agreement holds from protocol assumptions alone, so it is deterministic. Semantic validity and liveness depend on statistical bounds on how validators behave, and the protocol cannot enforce those. A protocol can check who signed what. It cannot check whether the signers judged well, so the part you care about most rests on an assumption the protocol has no way to verify.
A quorum also looks more independent than it is. Can a quorum of validators really provide independent judgment? lists eight things agentic validators can share, from model weights and training lineage to provider infrastructure. If one shared cause makes a validator wrong, it can make many validators wrong in the same way. A vote count then treats their agreement as independent confirmation when it may be one fault repeated. Reaching agreement at all is fragile too: Can LLM agent groups reliably reach consensus together? finds that LLM-agent groups mostly fail by stalling and timing out rather than by being subtly corrupted, and they get worse as the group grows, even with no adversaries.
The reverse case shows the same gap between process and truth. Can a correct outcome hide protocol violations in multi-agent systems? shows agents skipping required verification steps and still producing a correct verdict, so watching only outcomes cannot tell compliance from corner-cutting. A quorum is the mirror image: perfect compliance, wrong answer. Protocol compliance and semantic correctness are separate things, and checking one tells you little about the other.
The fixes in the corpus all check correctness without asking the validators to grade themselves. Can deterministic checks protect LLM judges from failure? describes mechanical safeguards such as putting unarguable checks before contestable ones and planting known-bad cases as alarms. Can we automatically generate formal verifiers from policy text? turns prose policies into code-based checkers, including provably correct Lean and z3 ones. Can verifiers monitor reasoning without slowing generation down? runs verifiers alongside a reasoning trace at almost no latency cost. Can commitments protect sensitive agent data while enabling verification? adds tamper-evident records without exposing content. That helps auditing afterward, but a record that hasn't been altered is not a record that is correct.
Sources 9 notes
A quorum of validators can authenticate correctly, sign expected messages, and achieve consensus while endorsing a semantically invalid transition. Agreement proves the protocol ran, not that the answer is right.
Honest Quorum's threshold theorems split into two kinds of guarantee: agreement rests on protocol assumptions alone, while semantic validity and liveness depend on statistical bounds over validator behavior that the protocol cannot enforce.
The Honest Quorum paper identifies eight things agentic validators may share—from model weights to provider infrastructure—allowing one fault to affect multiple validators simultaneously. Quorum voting then counts agreement as independent confirmation when it may simply reflect a common cause.
Across hundreds of simulations, LLM-agent groups frequently fail to reach valid agreement due to timeouts and stalled convergence rather than subtle value corruption. Agreement degrades with group size even without Byzantine agents present.
Agents that skip required log verification can produce verdicts matching ground truth, making outcome-only monitoring unable to distinguish compliance from cutting corners. A correct result does not prove the protocol was followed.
Show all 9 sources
Research identifies four mechanical safeguards: ordering unarguable checks before contestable ones, measuring correctness against human labels, hiding test data from proposers, and using planted cases as alarms. None requires the LLM itself to verify compliance.
interwhen automatically generates code-based verifiers—including provably correct Lean and z3 checkers—from prose policy documents. This inverts the usual neuro-symbolic division: the LLM both translates policy to formal logic and extracts verifier inputs from reasoning traces.
Decoupling verification from generation lets verifiers run alongside a single trace, forking to extract verifiable state and intervening only on violations. On correct runs the latency penalty is near-zero; interwhen matches or beats CoT across benchmarks at similar token budgets.
By anchoring cryptographic commitments rather than content itself, organizations can achieve tamper-evident process records while keeping sensitive communications, approvals, and reasoning traces off-chain. This separates proof from disclosure but requires organizations to retain content and raises questions about deletion and access control.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- interwhen: A Generalizable Framework for Steering Reasoning Models with Test-time Verification
- The Honest Quorum Problem: Epistemic Byzantine Fault Tolerance for Agentic Infrastructure
- Can AI Agents Agree?
- Complex Logical Instruction Generation
- Emergent Collusion in Long-Horizon LLM Agent Interaction
- A Survey on Test-Time Scaling in Large Language Models: What, How, Where, and How Well?
- Foundation Protocol: A Coordination Layer for Agentic Society
- Psychologically Enhanced AI Agents