SUBNET TRACKER · All subnets
SN60 · Bitsec.ai
Every Bittensor Daily mention of SN60 Bitsec.ai, newest first. Price changes are snapshots from each edition.
24h change −5.81%
- What happened
- A community member reported that the Initia Move project's H-02 vulnerability, which the SN60 (Bitsec.ai) benchmark is supposed to test for, has already been patched in the current benchmark image, meaning agents cannot legitimately detect it; the member requested a re-screening of agents evaluated after the report. Separately, miners asked whether the evaluation project list for a round is disclosed at submission time, and the team clarified that project keys are not known in advance and can change between rounds. A further discussion traced how one participant obtained a project list from hard-steer check logs, revealing that the project set had changed at the last moment before a round started.
- Community
- Several miners supported the request for re-screening, with one independently verifying the H-02 patch claim against the benchmark source code. Miners also pressed the team with detailed questions about when and how round-specific project lists become visible to agents, and disputed the source and timing of a leaked project list.
- Official
- A team member confirmed the H-02 vulnerability in Initia Move had been patched while the round still showed four errors, calling it an oversight. The team also confirmed that project keys are not disclosed in advance and can change between rounds, visible to miners only during evaluation.
24h change −5.39%
- What happened
- A miner publicly accused the previous round's winner and agent 5940 of circumventing SN60 (Bitsec.ai)'s hard-steering screener through mass submissions across multiple hotkeys, listing specific hotkeys and submission IDs as evidence. The agent's developer and other participants disputed the claim, arguing that the agent passed the platform's AST validation, hard-steering screener, and duplicate-fingerprinting checks, and that repeated submissions to refine an agent are permitted. The dispute extended over many messages without resolution in the channel.
- Community
- The exchange grew heated, including direct insults and calls to ban participants on both sides; moderators intervened to stop personal attacks and repeated pinging of the owner, asking that reports focus on specific exploit evidence. Separately, a participant proposed rotating benchmark projects each round since current targets appear to stay the same across rounds, a suggestion others in the channel agreed with.
- Official
- A moderator posted reminders asking participants to avoid personal attacks and repeated pinging of the owner, and to allow the team time to investigate reported issues.
24h change −5.55%
- What happened
- On SN60 (Bitsec.ai), several miners questioned the detection rates of top-scoring agents (5521, 5743, 5744), noting that passing stricter hard-steering criteria at such rates was, in their view, hard to explain, and asked the team to make the agent code public. One miner posted a breakdown arguing that prompt modules in agent 5521 closely matched descriptions of specific known project exploits. The team said it identified and fixed an issue in the hard-steering screener, after which a higher hard-steering score was observed.
- Community
- Reactions were split: some miners defended the top agents and found no major issues on review, while others repeated calls for the winning agent's code to be published and for the result to be revoked if objections surfaced. A proposal was raised to require public disclosure of all agent code going forward, with an added bonus for winners not challenged by other miners.
- Official
- The team stated the hard-steering screener issue was fixed, resulting in a higher score, and said any changes to evaluation criteria should apply to future competitions rather than be applied retroactively.