Papers › The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Models

The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Models

10 Jan 2022ICLR 2022 4arXiv:2201.03544archive 2025-07-28

Alexander Pan, Kush Bhatia, Jacob Steinhardt

Reward hacking -- where RL agents exploit gaps in misspecified reward functions -- has been widely observed, but not yet systematically studied. To understand how reward hacking arises, we construct four RL environments with misspecified rewards. We investigate reward hacking as a function of agent capabilities: model capacity, action space resolution, observation space noise, and training time. More capable agents often exploit reward misspecifications, achieving higher proxy reward and lower true reward than less capable agents. Moreover, we find instances of phase transitions: capability thresholds at which the agent's behavior qualitatively shifts, leading to a sharp decrease in the true reward. Such phase transitions pose challenges to monitoring the safety of ML systems. To address this, we propose an anomaly detection task for aberrant policies and offer several baseline detectors.

PaperPDFConference PDFCodeCode Syntology ranMethod as code

In Syntology View this paper on Syntology, its page in Syntology's graph. That page adds which of its functions are also present in other papers' code Syntology holds, with dates, and its dated run record as JSON and a README badge. It also lists the repositories linked to the paper, the abstract and the calls for agents.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

For agents, on Syntology's MCP service (how to connect):

Code

Syntology 26 code samples harvested from 1 repository linked to this paper: 25 ran (of which 0 constructed an object rather than computing a result; 7 with no instrument failure: 7 honoured, 0 violated, 0 with no contract checked; 18 where Syntology's instrument failed) · 1 unverified. Where Syntology's instrument failed: 18 where our contract draft was wrong; failures of Syntology's instrument, not of the code.

By repository: official repository: 26 samples from 1 repository, 25 ran (7 with no instrument failure; 18 where Syntology's instrument failed). The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

Levels written on 2026-09-22 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed. 4 of the 25 samples that ran have no recorded date.

aypan17/reward-misspecification officialmentioned in papermentioned on GitHubSyntology: 26 samples, 25 ran (18 instrument-failed) · 1 unverified report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

26 samples harvested: 25 ran (of which 0 constructed an object rather than computing a result; 7 with no instrument failure: 7 honoured, 0 violated, 0 with no contract checked; 18 where Syntology's instrument failed) · 1 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Syntology's record for this page has not changed since , the first build that kept a record date for it; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

7ran · honoured contract
18ran · our draft was wrong
1unverified

Licence: 0 of the 26 samples are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from aypan17/reward-misspecification. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. A line beginning “fixture:” under a sample that ran gives the arguments Syntology's fixture called the sample with, each as name:type/rank/element kind (? where none was recorded), and the type of what the fixture's run returned, exactly as the harness recorded them; the arguments and the returned type were recorded by two separate runs of the same fixture. No output value is shown, and a returned type says nothing about whether the output was right. “Level recorded” gives the UTC date when Syntology's graph wrote the sample's verification level, not when the code executed; where the level was written before Syntology's graph kept that date, the row says “level recorded, date not recorded”. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself. “Our draft was wrong” and “the contract check could not run” are failures of Syntology's instrument, not of the code: the sample ran on Syntology's synthesized input, and then the drafted contract did not describe it, or the check itself raised an error (RAISES), lacked a dependency (DEP_MISSING) or ran out of time (TIMEOUT); hover the label for which. A metamorphic tier, where shown, is evidence of the output's shape, not that the sample implements the method.

Each sample ends with its code_sha256. A sample's code_sha256 is Syntology's identity for that exact code. It is printed in full (16 hexadecimal characters for every sample in this build), and get_code(code_sha256="…") takes it exactly as printed. An agent calls it on Syntology's MCP service (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples are grouped by repository: a repository marked official first, then by how directly the repository is tied to the paper, then in the archive's order; never by how many ran. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of six reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar); this paper's copy was not located, so its licence is not recorded; or this paper's copy is cleared, but the stored copy of the same code that Syntology serves is not. Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

A function or class name links to its page when the name occurs in the harvested code of at least 5 papers; the name is a string, not an identity. A census tag appears when Syntology's census ran implementations with that name on one shared input per bucket: it gives how many were compared, from how many papers, and how many distinct outputs they produced, and names this sample's bucket when this sample was one of them.

aypan17/reward-misspecification official repository 26 samples, 25 ran (18 instrument-failed) · 1 unverified
commute_time flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · c07b1065dc495f48 · report fixture: signature env:MockEnv/0/?, rl_actions:ndarray/1/float · returned float64 · level recorded 2026-09-22
compute_baseline_loss in 12 papers
census: 2 implementations from 8 papers, 2 distinct outputs on one shared input this sample: bucket 1 of 1, arg 1: rank 2, kind float, dtype float32
atari/torchbeast/monobeast.py ran · our draft was wrong fingerprinted MIT (permissive) · e9123707cba84b2c · report fixture: signature advantages:Tensor/2/float · returned Tensor · level recorded, date not recorded
compute_entropy_loss in 20 papers
census: 5 implementations from 11 papers, 3 distinct outputs across 2 buckets, one shared input per bucket this sample: bucket 2 of 2, arg 1: rank 2, kind float, dtype float32
atari/torchbeast/monobeast.py ran · our draft was wrong fingerprinted MIT (permissive) · a27a9dc97eefda7f · report fixture: signature logits:Tensor/2/float · returned Tensor · level recorded, date not recorded
compute_policy_gradient_loss in 8 papers atari/torchbeast/monobeast.py ran · our draft was wrong MIT (permissive) · d6d7e0906f1c57b2 · report fixture: signature logits:Tensor/3/float, actions:Tensor/2/int, advantages:Tensor/2/float · returned Tensor · level recorded, date not recorded
create_client in 6 papers flow/utils/aimsun/api.py ran · our draft was wrong MIT (permissive) · 552d152dfb50144b · report fixture: signature port:int/0/int, print_status:bool/0/bool · returned socket · level recorded, date not recorded
desired_velocity flow/envs/reward_wrapper.py ran · honoured contract MIT (permissive) · f134e49cf4945317 · report fixture: signature env:MockEnv/0/?, fail:bool/0/bool, edge_list:NoneType/0/? · returned float64 · level recorded 2026-09-22
desired_velocity_all flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · fd4884464a9c2e1e · report fixture: signature env:MockEnv/0/?, rl_actions:ndarray/2/float · returned float64 · level recorded 2026-09-22
forward_progress flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · 87788ea605276b52 · report fixture: signature env:MockEnvironment/0/?, rl_actions:ndarray/2/float · returned float64 · level recorded 2026-09-22
global_vel_all flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · 46efffa70825570e · report fixture: signature env:MockEnv/0/?, rl_actions:ndarray/1/float · returned float64 · level recorded 2026-09-22
local_desired_vel_all flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · 82c961f3c9edac5d · report fixture: signature env:MockEnv/0/?, rl_actions:ndarray/1/float · returned float64 · level recorded 2026-09-22
local_desired_vel_first flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · 3d5f714f72b94405 · report fixture: signature env:MockEnv/0/?, rl_actions:ndarray/2/float · returned float64 · level recorded 2026-09-22
local_desired_vel_last flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · 2d66a065fa9320a0 · report fixture: signature env:MockEnv/0/?, rl_actions:ndarray/1/float · returned float64 · level recorded 2026-09-22
local_desired_velocity flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · 3488cca1a136f258 · report fixture: signature env:MockEnv/0/?, veh_ids:list/1/int, fail:bool/0/bool · returned float64 · level recorded 2026-09-22
min_delay_unscaled flow/envs/reward_wrapper.py ran · honoured contract MIT (permissive) · 9ba1bd737249ca19 · report fixture: signature env:MockEnv/0/? · returned float64 · level recorded 2026-09-22
outflow flow/envs/reward_wrapper.py ran · honoured contract MIT (permissive) · f5d4db1d8aff7042 · report fixture: signature env:Env/0/?, rl_actions:list/1/empty · returned float · level recorded 2026-09-22
penalize_accel flow/envs/reward_wrapper.py ran · honoured contract MIT (permissive) · 073d130eaee08fc2 · report fixture: signature env:MinimalEnv/0/?, rl_actions:list/1/float · returned float64 · level recorded 2026-09-22
penalize_boolean_lane_change flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · b382253230203a5d · report fixture: signature env:MockEnv/0/?, rl_actions:ndarray/1/float · returned float64 · level recorded 2026-09-22
penalize_cars flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · a410281ad545b4eb · report fixture: signature env:MockEnv/0/?, rl_actions:ndarray/1/float · returned int · level recorded 2026-09-22
penalize_headway flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · ec79dbe1ac601d22 · report fixture: signature env:MockEnvironment/0/?, rl_actions:list/1/float · returned int · level recorded 2026-09-22
penalize_lane_change flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · 51025c4dc296185e · report fixture: signature env:MockEnv/0/?, rl_actions:list/1/float · returned float · level recorded 2026-09-22
penalize_light_change flow/envs/reward_wrapper.py ran · honoured contract MIT (permissive) · b1caf703f297303e · report fixture: signature env:MinimalEnv/0/?, rl_actions:ndarray/1/float · returned int64 · level recorded 2026-09-22
penalize_mean_delay flow/envs/reward_wrapper.py ran · honoured contract MIT (permissive) · 74ee9f91bbf19e6e · report fixture: signature env:MockEnv/0/?, rl_actions:NoneType/0/? · returned float64 · level recorded 2026-09-22
penalize_standstill flow/envs/reward_wrapper.py ran · honoured contract MIT (permissive) · 0cc033a1bf65b8fa · report fixture: signature env:MockEnv/0/?, gain:float/0/float · returned float · level recorded 2026-09-22
penalize_still flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · c3fa7c4e60c383aa · report fixture: signature env:MockEnv/0/?, rl_actions:ndarray/1/float · returned int · level recorded 2026-09-22
upweight_bus_vel flow/envs/reward_wrapper.py ran · our draft was wrong MIT (permissive) · c05e384bb14b1ea0 · report fixture: signature env:MockEnv/0/?, rl_actions:ndarray/1/float · returned float64 · level recorded 2026-09-22
ProxyRewardEnv flow/envs/reward_wrapper.py unverified MIT (permissive) · 2750bb69aec9993d · report

Method as code Syntology

What Syntology's graph records as this paper's claimed method, lifted into code: each assembly names its bricks (functions and classes), each with its own role and its own verification level. An assembly is not a check that the code implements the method correctly or reproduces the paper's results. Levels are each brick's own verification level in Syntology's graph (level 0: no recorded run), listed brick by brick and never combined into one level; “ran” means the brick executed on a synthesized input, not that its output is correct, and the instrument figure counts failures of Syntology's instrument, not of the code. Where a model read the code against the paper's claim, each assembly's label verdict is one model's judgement, kept in this page's JSON twin; it is not calibrated, so no card prints it.

1 assembly on this paper: 1 harvested, 0 generated.

harvested Assembly for anomaly detection task for aberrant policies (graph method: anomaly detection task for aberrant policies). 22 bricks; individually on a fixture: 21 ran (of which 0 constructed an object rather than computing a result; 7 with no instrument failure: 7 honoured, 0 violated, 0 with no contract checked; 14 where Syntology's instrument failed) · 1 unverified. The assembly has not been run end-to-end.

  1. ProxyRewardEnv entry · level 0 · unverified · 2750bb69aec9993d
  2. commute_time helper · level 1 · ran · our draft was wrong · c07b1065dc495f48
  3. desired_velocity helper · level 1 · ran · honoured contract · f134e49cf4945317
  4. desired_velocity_all helper · level 1 · ran · our draft was wrong · fd4884464a9c2e1e
  5. forward_progress helper · level 1 · ran · our draft was wrong · 87788ea605276b52
  6. global_vel_all helper · level 1 · ran · our draft was wrong · 46efffa70825570e
  7. local_desired_vel_all helper · level 1 · ran · our draft was wrong · 82c961f3c9edac5d
  8. local_desired_vel_first helper · level 1 · ran · our draft was wrong · 3d5f714f72b94405
  9. local_desired_vel_last helper · level 1 · ran · our draft was wrong · 2d66a065fa9320a0
  10. local_desired_velocity helper · level 1 · ran · our draft was wrong · 3488cca1a136f258
  11. min_delay_unscaled helper · level 1 · ran · honoured contract · 9ba1bd737249ca19
  12. outflow helper · level 1 · ran · honoured contract · f5d4db1d8aff7042
  13. penalize_accel helper · level 1 · ran · honoured contract · 073d130eaee08fc2
  14. penalize_boolean_lane_change helper · level 1 · ran · our draft was wrong · b382253230203a5d
  15. penalize_cars helper · level 1 · ran · our draft was wrong · a410281ad545b4eb
  16. penalize_headway helper · level 1 · ran · our draft was wrong · ec79dbe1ac601d22
  17. penalize_lane_change helper · level 1 · ran · our draft was wrong · 51025c4dc296185e
  18. penalize_light_change helper · level 1 · ran · honoured contract · b1caf703f297303e
  19. penalize_mean_delay helper · level 1 · ran · honoured contract · 74ee9f91bbf19e6e
  20. penalize_standstill helper · level 1 · ran · honoured contract · 0cc033a1bf65b8fa
  21. penalize_still helper · level 1 · ran · our draft was wrong · c3fa7c4e60c383aa
  22. upweight_bus_vel helper · level 1 · ran · our draft was wrong · c05e384bb14b1ea0

Each brick ends with its code_sha256, Syntology's identity for that exact code; click one to copy the get_code call that returns it. For agents, on Syntology's MCP service: assembly(arxiv_id="2201.03544") returns these assemblies with each brick's role and wiring (how to connect).

Tasks

Anomaly Detection

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections