Two things worth remembering from July, even if monitoring is not your day job. First: the AI that watches over IT systems is no longer a demo, it is becoming a finished product, sold, documented, with support commitments. Second: several widely used monitoring tools received serious security fixes this month, and if your company runs one of them, that is the real priority. The rest, versions, conferences, weak signals, is sorted for you below.
In short:
- Autopilot, New Relic’s AI agent, reached general availability on 21 July. In other words, a piece of software that investigates production incidents on its own is now an official product at a major vendor, no longer an experiment.
- Centreon, SolarWinds and Paessler published security fixes, including two critical batches. No known attack to date, but if these tools run on your side, updating is this month’s priority.
- The rising question for IT leadership: governance of these AI agents. What scope of action, what traceability, which human stays in the loop when the agent gets it wrong.
The AI that watches your systems becomes a finished product
Let’s set the scene. A monitoring “AI agent” is a piece of software that observes your servers and applications, notices that an incident is brewing or has just broken out, investigates on its own by cross-checking the available data, then proposes a diagnosis and sometimes a fix. For a year now, every major vendor has been promising one. July’s news is the change of register: New Relic took its own, Autopilot, to general availability on 21 July. “General availability”, in vendor vocabulary, means the product leaves its trial phase: it is billed, supported, and the vendor stands behind how it works. Along the way, Autopilot now plugs into Slack, the team messaging tool: you can literally discuss an ongoing incident with it, in the very thread where the team is already organising.
In the same vein, New Relic presented Smart Alerts on 23 July, in public preview: an AI-strengthened alerting engine whose goal is to cut down the “noise”, those dozens of harmless notifications that end up drowning the one that matters. And to be complete on this vendor: Ground Truth, another product announced for “late July”, was still not officially available as we write. We will check again in August.
This movement is not limited to the American giants. Mr Suricate, a French automated-testing vendor, showed a feature in the same spirit in a 9 July webinar: when a test fails, the AI looks for the cause, checks its hypothesis by replaying the scenario in isolation, then proposes a documented fix that a human validates in one click. The vendor’s motto sums up the moment’s philosophy nicely: “AI prepares, humans decide.”
Why does this matter for IT leadership? Because once a product is generally available, teams will actually deploy it. And at that point the questions are no longer technical but about governance: what scope is the agent allowed to act on? Who reviews its conclusions? What happens when it gets things wrong? These questions now deserve written answers, just like an on-call procedure.
That same month, a wave of security fixes
July’s other defining fact is less cheerful but just as structural: within two weeks, three vendors published serious security fixes. A useful reminder to gauge the stakes: a monitoring tool sees your entire information system, that is its whole point. If it is compromised, the attacker inherits a detailed map of your IT and, often, privileged access. A security fix on these tools is therefore never a housekeeping detail.
The most serious case concerns Centreon: a critical flaw, rated 9.6 out of 10 on the CVSS scale that measures vulnerability severity, in the centreon-open-tickets ticketing module. In plain terms, a logged-in user could make the platform execute arbitrary code. The official bulletin documents the flaw (CVE-2026-14453) and its fix, shipped in versions 25.10.9 and 24.10.14 of the module. At SolarWinds, the Serv-U 2026.3 update of 21 July fixes sixteen vulnerabilities in one go on this file-transfer tool, several of them critical (up to 9.1), mostly access-control flaws. Finally at Paessler, the maker of PRTG, a coordinated disclosure published on 9 July details six vulnerabilities of more moderate severity (6.8 at most). “Coordinated disclosure” means researchers reported the flaws to the vendor, the fix shipped first, back in early June, and publication only came afterwards: that is the virtuous sequence.
A reassuring note: in all three cases, no active exploitation has been observed to date. And a positive signal in the same vein: VictoriaMetrics tightened the permissions of its Kubernetes operator, the component that drives its deployment, which had broader default rights than necessary. The version carrying that hardening introduced a regression; it was fixed within twenty-four hours. A vendor that repairs its own mistakes in a day is exactly the reflex you want to see.
The month’s lesson fits in one sentence: the pace at which a vendor fixes things remains the first indicator of its maturity, ahead of its list of AI features.
On routine maintenance, the machine keeps turning
For the rest, July was a month of ordinary maintenance, and that is good news: tools that receive small regular updates are living tools. Zabbix shipped three corrective releases, Centreon kept up its fix cadence and enriched its connector catalogue, the building blocks that let you monitor a given device or application, and VictoriaMetrics published two versions, with a single trap to know about: the first (v1.147.0) suffers a performance regression in some configurations, and the official guidance is to jump straight to v1.148.0. If a service provider runs your monitoring, keeping up with all this is their job; the full version rundown is at the end of this article for technical teams.
Also worth noting
SolarWinds was reconfirmed for the second year running in the 2026 Gartner Magic Quadrant for observability platforms, the sector’s reference ranking, and extended its database analysis tool DPA to SAP HANA Cloud. New Relic announced its RUNTIME user conference, 12 and 13 August in Atlanta. IP-Label published a reflection on sustainable computing in late July: decades of energy-efficiency gains in data centers, and AI now putting those gains into question. Finally, in the interest of transparency: M81 and Martello announced nothing in July. For IBM i estates monitored with M81, note all the same that IBM shipped its annual technology refresh mid-July (7.6 TR2 and 7.5 TR8).
What to remember
- Operational AI is becoming a product you buy, and therefore a governance topic. Who is accountable when the agent gets it wrong: that question deserves a written answer before deployment, not after the first incident.
- This month’s priority is defensive. If Centreon, Serv-U or PRTG run on your side, check that July’s fixes (and June’s for PRTG) are applied. Everything else can wait a week; not that.
- Judge your vendors on their fix cadence. A regression corrected within twenty-four hours says more than a list of AI promises.
Observing the agents: our two in-house bricks, SenHub agent and Toise
With AI agents now sold as finished products, one question becomes very concrete: when an agent changes something on your infrastructure, how do you know what it touched and what that impacts? This is the conviction behind our two open-source projects, SenHub agent and Toise, both natively compatible with OpenTelemetry, the open standard of observability.
SenHub agent collects monitoring data, infrastructure and databases, in a standardised and portable way: your monitoring data stays yours, with no proprietary lock-in.
Toise maps the entities of your information system and their dependencies, with their history, and answers the impact question: “if this component goes down, or if an AI agent changes it, what does it touch?” That is precisely the traceability AI-agent governance needs, and Toise aligns with the sector’s emerging standard, OpenTelemetry’s “Resources and Entities” work.
What we do with it at Sensor Factory
Concretely, this July shifts our priorities toward immediate patching: Centreon and SolarWinds fixes handled as urgent for affected clients, PRTG version checks, and steering around the flawed VictoriaMetrics release. And with clients considering an operational AI agent, we are documenting a simple governance framework: scope of action, traceability, human in the loop.
If you want to take stock of your own stack, priority fixes or AI-agent governance, let’s talk.
For technical teams: July’s version rundown
- Zabbix: 7.0.28 LTS (7 July), 7.4.12 (8 July), 7.4.13 (29 July). Nothing disruptive, 7.0 remains the recommended target for new deployments.
- VictoriaMetrics: v1.147.0 (6 July:
default_vm_access_claimfield on vmauth, roughly 10% less CPU on remote-storage sharding; CPU/memory regression with-remoteWrite.urlRelabelConfigor-remoteWrite.streamAggr.config, avoid it), v1.148.0 (20 July: “fill” modifiers in MetricsQL, Unix-domain-socket scraping,-maxBackfillAgeflag); Kubernetes operator v0.73.0 (7 July, namespace-scoped secrets/configmaps permissions) and v0.73.1 (8 July, fixes the previous day’s VMRule regression). - Centreon:
centreon-collect-24.10.22(10 July),25.10.10(16 July),25.10.11+monitoring-agent-25.10.8(22 July),24.10.23+monitoring-agent-24.10.18(28 July) on GitHub; plugins-20260700 catalogue (22 July: new Waystream connector, improvements to 3CX, Cisco Meraki, F5OS, Oracle Database, Palo Alto, Proxmox VE, VMware8, fixes for Commvault, MS SQL Server, Veeam); Open Tickets 25.10.9 / 24.10.14 (CVE-2026-14453 fix). - SolarWinds: Serv-U 2026.3 (21 July, 16 CVEs fixed, OpenSSL 3.0.21), DPA 2026.2 (9 July, SAP HANA Cloud support, AI-assisted tuning).
- Paessler: coordinated disclosure of 9 July (2 XSS, 3 information disclosures, 1 timing-based authentication weakness, CVSS at most 6.8), fixed in PRTG 26.2.120.1449, shipped in early June.
- New Relic: Autopilot general availability + Slack integration (21 July), Smart Alerts in public preview (23 July).
This tech-watch is a monthly format. The next edition will cover August 2026. To receive it automatically, the newsletter sign-up is at the bottom of the page.
