Vicarius puts a human reviewer and a deterministic judge between its AI-written fixes and production

An advisory lands at four in the afternoon. The affected software is in your estate, the vendor says a fix is coming, and nobody will say when. Until it arrives, the choice is a workaround somebody on your team writes tonight or nothing at all. Vicarius is now selling the first option pre-drafted, written by a model and checked by its own researchers, inside the vRx platform its customers already run ^[1][5]. The other no-patch paths in that platform are vShield, which protects the workload, and vScript, which holds custom and community-sourced scripts ^[17].

The urgency case rests on a number Vicarius did not produce. Zero Day Clock data cited this month by Radware put mean time to exploit at negative eight hours as of July, against 53 days for the same measure in 2024 ^[3][4]. Fifty-three days is 1,272 hours, so the figure has moved 1,280 hours in two years ^[19]. A negative mean puts the exploitation ahead of the disclosure that is supposed to warn you ^[21].

"Finding ten times more vulnerabilities is meaningless if remediation still happens through tickets, manual research and weeks of engineering work," said Roi Cohen, co-founder and chief executive of Vicarius ^[13]. The company names Anthropic's Claude Mythos and Google's Big Sleep agent as evidence of how fast machines now surface exploitable code, and says the fixes still have to be written ^[16].

The checking is what Vicarius added since its last run at this. The engine drafts detection and remediation logic, a Vicarius researcher reviews it, and the platform then executes it and verifies the result on affected machines ^[7]. A deterministic judge issues the pass or fail, with no language model in the decision path, the company said ^[8]. Each validated case ships with a cryptographically signed evidence manifest that customers can verify themselves ^[9]. The script runs a pre-check, applies the fix, verifies it, and rolls back on failure ^[10].

vuln_GPT arrived in August 2023, billed as the first model trained specifically to write remediation scripts, and its output went out free through the company's vsociety community ^[12]. Vicarius says the new sequence takes under an hour, against weeks for the same work done by hand ^[6]. That figure is the company's own, and the launch report does not include pricing or any customer results ^[20].

Two questions decide whether any of this moves your time to remediate. What do you do in that window today, and who is allowed to approve a script that touches production? If the honest answer to the first is nothing, a reviewed script with a rollback beats nothing by however often the reviewers are right, and you will not have a baseline to check them against. If your change process needs a named human owner for every script that runs with admin rights, drafting was never your slow step, and an hour of generation will sit a week waiting for a window. The scripts are native Bash and PowerShell at the operating system level with no extra agent and no proprietary rule format ^[11], so the approval they need is the approval any script already needs.