Autonomous AI Security Research
Assurance Infrastructure
Making autonomous AI agents trustworthy enough to rely on.
Auron Lab is a security research labs building the assurance and evaluation infrastructure for AI agents. Our work supports safer adoption of AI in software workflows where security is critical.
See what we are building
tempo
/
reth-execution-engine
Link GitHub Repo
Active
File browser
crates
evm
src
lib.rs
execute.rs
processor.rs
payload
src
lib.rs
builder.rs
transaction-pool
src
lib.rs
pool.rs
validator.rs
precompiles
src
lib.rs
primitives
src
lib.rs
consensus
src
lib.rs
commonware-node
src
lib.rs
execute.rs
processor.rs
builder.rs
1
use crate::TempoEvmConfig;
2
use alloy_consensus::crypto::RecoveryError;
3
use alloy_primitives::Address;
4
use rayon::prelude::*;
5
6
ConfigureEngineEvm, ConfigureEvm, EvmEnvFor, ExecutableTxIterator, ExecutionCtxFor,
7
FromRecoveredTx, RecoveredTx, ToTxEnv, block::ExecutableTxParts,
8
};
9
use reth_primitives_traits:: {SealedBlock, SignedTransaction};
10
use std::sync::Arc;
11
use tempo_payload_types::TempoExecutionData;
12
use tempo_primitives::{Block, TempoTxEnvelope};
13
use tempo_revm::TempoTxEnv;
14
15
impl ConfigureEngineEvm > TempoExecutionData > for TempoEvmConfig {
16
fn evm_env_for_payload(
17
&self,
18
payload: &TempoExecutionData;
19
) -> Result<EvmEnvFor<Self>, Self::Error> {
20
self.evm_env(&payload.block)
21
}
22
23
fn context_for_payload<'a>(
24
&self,
25
payload: &'a TempoExecutionData,
26
) -> Result<ExecutionCtxFor<'a, Self>, Self::Error> {
27
let TempoExecutionData { block, validator_set } = payload;
28
let mut context = self.context_for_block(block)?;
29
30
context.validator_set = validator_set.clone();
31
32
Ok(context)
33
}
34
35
fn tx_iterator_for_payload(
36
&self,
37
payload: &TempoExecutionData,
38
) -> Result<impl ExecutableTxIterator<Self>, Self::Error> {
39
let block = payload.block.clone();
All threads
Tempo/reth-execution-engine
Open
Nathaniel
Auron triage Team
19:20
A thread for project-level discussions and important notifications.
Auron
19:20
@Protocol team New finding available:
H-01: Pre-T2 Fee Swap DoS: Fee AMM liquidity can be drained mid-tx to force collect_fee_post_tx() to error as EVMError::Custom
Horsefacts
Protocol team
19:20
Thank you! Great findings so far, will review this now.
I think this one is quite significant too.
Auron
19:20
@Protocol team New finding available:
H-02: StablecoinDEX pre-T1A swallowed OutOfGas during flip placement commits partial state
Georgios
Protocol team
19:20
These are some nice finds.
Your message here...
WHAT WE DO
We making AI security agents, measurable, comparable and safe to use.
Autonomous agents are moving from code completion into code analysis, vulnerability discovery, remediation and security reasoning. Before organisations rely on them, they need to know whether AI agents are reliable in software workflows where security is non-negotiable. Today, that is still difficult to test consistently.
Tested on real world data
AI security agents are evaluated against real world vulnerabilities and working exploits in safe and controlled sandboxes.
Reproducible and comparable
The same tasks, environments and scoring across agents, so results can be reproduced, trusted and compared.
Shared assurance infrastructure
Curated datasets, reproducible testbeds, evaluation harnesses and scoring methods for AI security agents.
ASSURANCE INFRASTRUCTURE IN DEVELOPMENT
The Assurance Asset/ Approach.
The asset is designed as a data-and-evaluation pipeline. It starts with vulnerability inputs from public advisory sources, then verifies, de-duplicates and documents them before any record becomes part of an evaluation task. Accepted records are linked to source, provenance, affected versions, fixes, validation status, environment manifests and known limitations.
Curated data records
Reproducible environments
Agent-facing harness
Security-aware scoring
Governed access
WHO BENEFITS
Built for the UK AI security ecosystem.
Register interest in
Assurance infrastructure.
Auron Labs is accepting a limited number of teams before the public launch. Join the waitlist and be first to know when early access opens.