Bitsec (SN60) Just Hit a New Standard for AI Security Agents

A security agent that finds every critical and high-severity vulnerability in a codebase, in under 30 minutes, running on open-source models, is the kind of output that should make a serious security team look twice.

Bitsec (Subnet 60) is producing exactly that this round, with the top agent clearing the bar on more than half of all projects tested. The v3.1 release that made it possible bundled execution-time extensions, automatic emissions transitions, downloadable run data, and a redesigned leaderboard into a single coordinated push.

The combination pulled 47+ agents into the round and gave the subnet its sharpest performance numbers to date.

What the Agents Are Doing

The performance numbers from this round are worth pulling out individually, because they reframe what Bitsec’s miner network is currently capable of:

a. 47+ agents participated in the most recent round, a meaningful jump driven directly by the v3.1 changes.

b. The top agent scored above 50% on the subnet’s evaluation suite.

c. Top agents find every critical and high-severity vulnerability in more than half the projects they are tested against, using open-source models.

d. The full run completes in under 30 minutes, which is the kind of latency that matters for any team thinking about plugging this into a real security workflow.

That combination, open-source models, sub-30-minute runtime, and 100% critical-vulnerability recall on half the test set, is what makes the subnet interesting beyond its own emissions story.

The v3.1 Changes That Moved the Numbers

The update split into three categories, each addressing a different reason miners were not engaging at scale before. 

The mechanism improvements:

a. Sandbox execution time extended from 20 to 30 minutes, giving agents room to handle larger projects without timing out.

b. Automatic emissions switch-over moves rewards to the new round’s winner immediately after evaluation closes, removing manual delay between rounds.

c. Tighter scoring logic with better tie-breaks and a visible “projects passed” metric, making the ranking system more transparent.

The visibility and feedback improvements:

a. Live subnet snapshot on the leaderboard showing round status, submissions, close dates, evaluated counts, recent activity, project counts, and emissions status.

b. Agent detail pages with clearer score summaries, validator breakdowns, project-level results, passes, true positives, and evaluation status.

c. Downloadable .JSON exports of agent runs from the detail page, giving miners direct access to execution and evaluation data including matched and missed findings.

The performance and UX (User Experience) upgrades:

a. Backend views, indexes, and agent detail endpoints optimized for faster, more stable performance during active evaluations.

b. Refined UI (User Interface) across agent details, validator cards, problem results, loading states, and empty states for a cleaner review experience.

The pattern across all three categories is the same: give miners more time, more transparency, and more data so they can iterate on their agents without fighting the platform.

Where This Lands

Bitsec’s v3.1 is the kind of update most subnets need but rarely ship cleanly. The mechanism improvements removed friction from the mining loop, the visibility upgrades gave operators real feedback on what their agents are doing, and the data exports turned the subnet into something miners can actually train against.

The 47-agent round and the 50%-plus top score are the early signal that the changes are working, and the team has already flagged that brand new challenge problems are coming in the next round with further incentive mechanism tuning aimed at revenue-generating customers.

The trajectory from here matters because Bitsec is one of the cleaner examples of a subnet producing outputs that map directly to a real enterprise need, and the v3.1 changes look like the right groundwork for what comes next.