Superpower Daily
Abliteration Removes Refusal Mechanisms for Offensive Cyber WorkLlamaIndex and Kaggle Launch a Document Benchmark for Cost and...Mostik Links AI Models Through Their Weights, Claiming a 20x Co...Lyte Raises $165M at $1.6B Valuation for Robot PerceptionNew Mexico Lets Districts Seek Approval to Leave K-2 AI Reading...HPE Raises 2026 Revenue Growth Forecast to 34%-37% After $12.21...Abliteration Removes Refusal Mechanisms for Offensive Cyber WorkLlamaIndex and Kaggle Launch a Document Benchmark for Cost and...Mostik Links AI Models Through Their Weights, Claiming a 20x Co...Lyte Raises $165M at $1.6B Valuation for Robot PerceptionNew Mexico Lets Districts Seek Approval to Leave K-2 AI Reading...HPE Raises 2026 Revenue Growth Forecast to 34%-37% After $12.21...
Toolspublished

Capsule Security Adds a Last-Second Block for AI-Agent Actions

The new control layer judges a planned action before it runs, rather than relying only on access permissions or post-incident review. Its reported benchmark results are promising, but the company’s internal model comparison remains difficult to assess fully.

By 2 min read
Capsule Security Adds a Last-Second Block for AI-Agent Actions
Capsule Security Adds a Last-Second Block for AI-Agent Actions
Sign in to save and react

Listen to this story

The audio brief

About 1:37
0:001:37
Read transcript
Capsule Security has released a control layer that can block an AI agent’s next action at the last second, just before that action executes. The system evaluates a specific planned step and gives customers three choices: allow it, flag it, or block it. That creates a decision point between broad permissions and after-the-fact monitoring. An agent might technically have access to code, sensitive data, or infrastructure, while a particular operation still falls outside its assigned task. Capsule’s approach is designed to catch that mismatch before the operation lands, rather than after damage has occurred. The company says its detectors classify actions instead of generating full responses, helping keep the check fast enough for the execution path. Capsule reports decisions in as little as 71 milliseconds. It trained the system on two fine-tuned Nvidia Nemotron models, using agent traces, human review, and adversarial examples. Capsule also says it nearly halved the larger model’s memory requirement without a reported performance loss, allowing deployment on one Nvidia L40S GPU. On StepShield, a benchmark built from 9,429 code-agent trajectories linked to real incidents, Capsule reports 98 percent accuracy. Its internal comparison showed 96.9 percent for its detector versus 86 percent for an unnamed third-party model. That result is harder to judge independently because the model and score breakdown were not disclosed. The key constraint is clear: the runtime control is available now, but the strongest comparative evidence remains Capsule’s own.

Story brief

3 key points

Capsule Security is now offering a runtime guardrail that judges an agent’s next step just before it executes, with customer-configurable allow, flag, or block outcomes. The system targets actions involving code, sensitive data, and infrastructure—areas where static permissions may miss context and monitoring arrives too late. Capsule reports decisions in as little as 71ms, deployment on one Nvidia L40S GPU, and 98%...

  1. 01

    Capsule says its detectors classify planned actions rather than generate responses, helping keep latency low enough for execution-path controls.

  2. 02

    Training used agent traces, human review, and adversarial examples with two fine-tuned Nvidia Nemotron models.

  3. 03

    Memory use for the larger model was nearly halved without reported performance loss, enabling single-GPU deployment.

Capsule Security has released a runtime control layer that checks an AI agent’s intended action immediately before execution. Customers can allow, flag or block the step, creating a decision point for agents that can access sensitive data, write code or operate infrastructure.

A check before the action lands

The capability is available now. Capsule places the control outside the agent, where it evaluates a specific planned step rather than simply setting broad access rights. The company says permissions and approval workflows can restrict what an agent may touch, but may not determine whether a particular action fits its assigned task. Post-incident monitoring, by contrast, finds a problem after damage has occurred.

That is the basis for Capsule’s circuit-breaker framing. Its detectors perform a narrow classification task instead of generating a full response, a design the company says lets them sit in an agent’s execution path with little delay.

Two models trained for one judgment

Capsule built the detection system on two Nvidia Nemotron models fine-tuned for real-time action evaluation. Nvidia Nemotron 3 Ultra supported the training run, which combined agent traces, human review and adversarial examples.

  • Agent traces
  • Human review
  • Adversarial examples

Capsule says decisions arrived in as little as 71 milliseconds. It also says it cut the larger model’s memory requirement by nearly half without a performance hit, allowing deployment on a single Nvidia L40S GPU.

The comparison still needs context

Capsule said the system reached 98% accuracy on StepShield, an academic benchmark containing 9,429 code-agent trajectories drawn from real incidents. Its strongest detector scored 96.9% in Capsule’s internal testing, compared with 86% for the strongest third-party general-purpose model it evaluated.

The two results answer different questions. Capsule did not identify the third-party model or provide a score breakdown for its internal comparison, limiting how fully that result can be assessed. Capsule also says billions of tokens across millions of agent interactions pass through its technology, with customers in financial services and technology.

Sources

  1. markets.businessinsider.comCapsule Security Builds “AI Circuit Breaker” for Rogue Agents with NVIDIA Nemotron
  2. siliconangle.comCapsule Security fine-tunes Nvidia Nemotron models to stop rogue AI agents - SiliconANGLE

Related coverage

Continue reading