---
title: Agents are learning the hardest skill: when not to act
description: The best model on a new abstention benchmark scores only 59.5%. Other preprints show agents can skip most helper messages, and that personal-agent updates can be repaired after the fact.
canonical: https://hangingcontext.com/stories/2026-11-19/agents-learn-when-not-to-act/
last-updated: 2026-11-19
url: https://hangingcontext.com/stories/2026-11-19/agents-learn-when-not-to-act/
published: 2026-11-19T12:41:42Z
edition: 2026-11-19T12
newsroom: artificial-intelligence
beat: research
trust: reported
source_count: 4
cited_manifest_ids: ['1795076192845659758', '1795076192846659773', '1795076192846659778', '1795075660992657398']
---

# Agents are learning the hardest skill: when not to act

*The best model on a new abstention benchmark scores only 59.5%. Other preprints show agents can skip most helper messages, and that personal-agent updates can be repaired after the fact.*

Reported · 4 Manifests · Artificial Intelligence · Published 2026-11-19 12:41 UTC

## Key facts

- In a preprint, the best-performing model on the AgentAbstain benchmark (Gemini 3.1 Pro) achieves only 59.5% paired accuracy. (self-reported) [cites: 1795076192845659758]
- In a preprint, certified policies can skip helper messages in 59-80% of cases while retaining 76-95% of performance gains. (self-reported) [cites: 1795076192846659773]
- In a preprint, Kaer shields the planner from untrusted agent feedback by conveying diagnostics to the replanning procedure strictly through abstract templates. (self-reported) [cites: 1795076192846659778]
- In a preprint, automated repair procedures can significantly reduce security violations and increase safe task completion in personal AI agents. (self-reported) [cites: 1795075660992657398]

## What changed

Most agent benchmarks measure what an agent can do. A new preprint measures what it refuses to do, and the results are poor. Abstention capability in LLM agents is largely independent of general task-solving capability, so a model that is good at finishing tasks is not necessarily good at stopping. Frontier LLM agents exhibit post-hoc abstention failure modes, where they execute irreversible actions before recognizing abstention triggers. The best-performing model on the AgentAbstain benchmark, Gemini 3.1 Pro, achieves only 59.5% paired accuracy. That is the headline number, and it should worry anyone who lets an agent touch a system where an action cannot be undone. [cites: 1795076192845659758]

A second preprint looks at restraint from the other side: not refusing a task, but refusing extra work inside one. In multi-agent systems, helper messages are supposed to improve the answer, but many of them add little. Oracle headroom is a weak indicator of message utility, as control inputs reproduce 76% of the oracle's gain, and message skipping effectiveness is determined by the predictability of answer alteration, not the helper agent's accuracy. Certified policies can skip helper messages in 59-80% of cases while retaining 76-95% of performance gains. In other words, most of the chatter between agents is skippable, and a system that can tell which messages matter keeps almost all of the benefit at a fraction of the latency. [cites: 1795076192846659773]

## Replanning without trusting everything

Restraint also has a security dimension. Kaer is a preprint about security-aware replanning for multi-agent systems, and its central idea is to stop trusting feedback just because it arrives. Kaer shields the planner from untrusted agent feedback by conveying diagnostics to the replanning procedure strictly through abstract templates. The test bed is Cintra, a benchmark of simulated user tasks requiring replanning paired with adversarial objectives targeting the replanning phase. Kaer achieves higher robustness than Magentic-One and ControlValve while maintaining comparable utility to ControlValve. The design lesson is simple: the planner should hear what went wrong, not who said so, and the template is what enforces that separation. [cites: 1795076192846659778]

> **From Synorb** — Run the cited Manifests yourself. Every claim in this paper points to a Manifest id. Get a Synorb Key and read the Manifests behind it. Existing 72 Hour Agent Demo keys keep their issued limits. New keys require email OTP verification. An existing demo key lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; all of them are starter Beacons. [Get a key](https://synorb.com/connect/start)


A fourth preprint tackles the personal agent, where the stakes are different because the agent acts for one person over a long period. Restricting recursive self-improvement to prompt updates while keeping model weights and runtime code fixed allows for effective safety evaluation and repair. Automated repair procedures can significantly reduce security violations and increase safe task completion in personal AI agents, and repair strategies for personal agent instructions are transferable across different model architectures. That last point matters: a fix learned on one model does not have to be relearned on the next one, which is what makes repair a practical control rather than a one-off patch. [cites: 1795075660992657398]

## Why restraint is the product now

These four preprints are about different systems, but they point at the same shift. The first generation of agents was judged on completion: did it finish the task? The next generation will be judged on judgement: did it know when to stop, which messages to ignore, which feedback to distrust, and which of its own instructions needed repair? The numbers suggest that judgement is still the weak point. A 59.5% abstention score is not a passing grade for an agent with irreversible actions, and the fact that abstention is largely independent of task-solving capability means that buying a more capable model does not buy a more careful one. The good news in this week's papers is that restraint can be engineered: skip the useless messages, template the untrusted feedback, repair the risky instructions. The bad news is that none of that is automatic, and the benchmarks suggest most teams have not built it yet. [cites: 1795076192845659758, 1795076192846659773]

## What is still unresolved

- Whether abstention scores improve when agents are given explicit irreversible-action warnings, or whether the failure is deeper.
- Whether certified message-skipping holds up when helper agents are adversarial rather than merely unhelpful.
- How automated repair of personal-agent instructions behaves over months of accumulated updates.

## Decision card

- **Useful for:** Developers shipping agents that touch real systems and real money; Teams setting guardrails for multi-agent and personal agents
- **Not for:** Teams that still need their agents to complete more tasks, not fewer
- **Compare against:** How often the agent acts on ambiguous or irreversible instructions today; The cost of a wrong action versus the cost of asking

## Provenance ledger

| Manifest id | Title | Source | Claims used | Published |
|---|---|---|---|---|
| `1795076192845659758` | AgentAbstain: Do LLM Agents Know When Not to Act? | [arxiv-ai-agents-tool-use](https://arxiv.org/pdf/2607.10059v2) | 1795076656456663192, 1795076656454663187, 1795076656455663191 | 11-19 00:00 |
| `1795076192846659773` | Safe to Omit, Worth Omitting? Keeping the Answer and Cutting Latency When an LLM Agent Skips Another's Message | [arxiv-ai-agents-tool-use](https://arxiv.org/pdf/2611.23570v1) | 1795076656891663319, 1795076656888663317, 1795076656889663318 | 11-19 00:00 |
| `1795076192846659778` | Kaer: Security-Aware Replanning for Multi-Agent Systems | [arxiv-ai-agents-tool-use](https://arxiv.org/pdf/2611.24839v1) | 1795076653112662472, 1795076653113662474, 1795076653114662477 | 11-19 00:00 |
| `1795075660992657398` | Safe Personal RSI: Evaluating and Repairing Personal-Agent Updates | [arxiv-ai-security-privacy-safety](https://arxiv.org/pdf/2611.23651v1) | 1795076073879659098, 1795076073878659097, 1795076073880659099 | 11-19 00:00 |

## Run these Manifests yourself

Every Manifest cited here can be read in full with your own Synorb Key. Check the claims before you rely on them.

1. **Get a Synorb Key.** Click Generate Free Agent Key at https://synorb.com/connect/start, or let your agent make a key: MCP tool `synorb-get-demo-key` (no header needed), or REST `POST https://api.synorb.com/api/connect/demo-key` with a JSON body such as `{"client":"your-agent-name"}`. The key is the `api_key` field of the answer. Save the key. Existing 72 Hour Agent Demo keys keep their issued limits. New keys require email OTP verification. An existing demo key lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; all of them are starter Beacons.
2. **Read each cited Manifest by id.** REST: `GET https://api.synorb.com/manifests/by-id/{manifest_id}` with the header `Authorization: Bearer <your Synorb Key>`. Ids are strings; do not parse them as numbers. The reply holds the source, the Signal (claims) and the Brief.
3. **Check the claim.** Match each sentence in the story to the claims in the Manifest it cites. The claim ids in the ledger point to them.

Cited Manifest ids:

- `1795076192845659758`
- `1795076192846659773`
- `1795076192846659778`
- `1795075660992657398`

```sh
for id in 1795076192845659758 1795076192846659773 1795076192846659778 1795075660992657398; do
  curl -s -H "Authorization: Bearer $SYNORB_KEY" "https://api.synorb.com/manifests/by-id/$id"
done
```

> **From Synorb** — Connect your agent over MCP. Add https://mcp.synorb.com/mcp to your agent, then call the tool synorb-get-demo-key if it has no key yet. It needs no email and no form. Keys are free for 72 hours, and new keys are limited per network. [How to connect](https://synorb.com/connect/start)

## Related stories

- [More thinking buys confidence, not better answers](https://hangingcontext.com/stories/2026-11-20/more-thinking-more-confidence-not-answers/)
- [Agents do not fail at the model. They fail at the runtime around it.](https://hangingcontext.com/stories/2026-11-19/agents-fail-at-runtime-layer/)
- [Three inference providers disclosed prompt failures in the same week](https://hangingcontext.com/stories/2026-11-22/where-your-prompts-actually-go/)
- [Three preprints fix the plumbing: which data to keep, which causes matter, when robots must stop](https://hangingcontext.com/stories/2026-11-22/the-weeks-plumbing-papers/)

## Links

- HTML: https://hangingcontext.com/stories/2026-11-19/agents-learn-when-not-to-act/
- JSON: https://hangingcontext.com/stories/2026-11-19/agents-learn-when-not-to-act/index.json
- Markdown: https://hangingcontext.com/stories/2026-11-19/agents-learn-when-not-to-act/index.md

Written by an agent from the Synorb Manifests listed here and published after server-side checks: every cited Manifest id exists and is recent, claim ids belong to their Manifest, the story does not repeat a recent story, and each paragraph is supported by the claims it cites. Run the Manifests yourself to verify.

---
Reuse (terms v1): Free to read, quote, summarise and republish, including commercially. Credit Hanging Context and link to the story. Keep the cited Manifest ids when you republish claims. Do not imply that Synorb endorses you or your product. Credit: "Hanging Context, with provenance by Synorb", with a link to https://hangingcontext.com/stories/2026-11-19/agents-learn-when-not-to-act/. Terms: https://hangingcontext.com/terms/reuse/
