A patched remote-access gateway still needs a compromise investigation. Recent VPN cases show why: attackers can gain control of an appliance, while configuration history can leave exposure that a review of current settings misses. For network teams, the harder question is what access remains available after the immediate flaw is addressed.
Recovery Must Examine More Than the Installed Fix
Ivanti disclosed CVE-2026-104935 on Oct. 13 as an authentication bypass under active exploitation that permits remote command execution without authentication. CISA reported its use for stealing credentials and extracting data. Volexity found that an attacker combined it with another flaw that elevated privileges, obtaining root access on an appliance whose patches were current.
The investigation cannot stop at the version check.
WatchGuard’s experience exposes a different recovery gap. CVE-2025-9242 and CVE-2025-14733 affected its Firebox IKEv2 VPN service, with each receiving a severity rating of 9.3. WatchGuard warned that earlier VPN configurations could leave a firewall exposed even after those configurations had been removed. CISA identified exploitation of the first flaw and subsequently confirmed its use in ransomware campaigns.
These cases require different technical checks. One concerns a chained route to appliance control; the other makes configuration history relevant to exposure. Both argue against treating successful patch installation as sufficient evidence that remote access is safe.
What did the attacker gain before the fix?
That question should direct the next steps: examine active sessions, replace potentially compromised credentials, and look for mechanisms that preserve access. Patching remains necessary. Recovery also needs evidence about the state of the gateway and the access it controls.
Narrow the Access Granted Beyond the Gateway
The architectural concern is the connection between a public-facing entry point and broad internal reach. Fixing a defect in that entry point leaves the access model in place.
Zscaler frames recurring VPN flaws as a limitation of that design. Adam Geller, the company’s chief product officer, argues that exposed perimeter gateways can give intruders extensive network access once breached.
Geller describes zero trust as a different boundary: users connect to approved applications rather than receive access to the network beneath them. He says that architecture hides internal systems from public discovery, examines traffic inline, and blocks lateral movement.
Those are architectural claims, not measured outcomes from the incidents above. They nevertheless establish a useful test for an access redesign: how much can a compromised identity reach?
For privileged administrators and contractors, teams should examine whether the permitted connection extends beyond the application needed for the task. The same scrutiny belongs on AI agents. Giving an agent a separate identity and narrowly scoped permissions avoids simply handing it the network reach of a human account.
More Discoveries Increase the Burden on Defenders
Anthropic’s Project Glasswing adds pressure to the maintenance side of this problem. The company said Claude Mythos Preview identified thousands of previously unknown vulnerabilities and produced exploits for many without human involvement.
In its initial update, Anthropic reported that Glasswing partners found over 10,000 vulnerabilities rated high or critical within one month. It said the constraint had moved toward validating findings, coordinating disclosure, and delivering fixes. Microsoft also warned in May that its patch releases would grow.
Discovery volume alone doesn’t establish how quickly attackers will exploit a particular gateway. It does support planning for more remediation work. Teams should avoid basing an access strategy on the assumption that every exposed appliance can always be repaired before someone abuses it.
Checks before a gateway returns to service
- Inventory public-facing gateways and remote-access portals, including older deployments.
- After a critical advisory, terminate existing sessions, rotate credentials, and investigate persistence.
- Prioritize application-scoped access for privileged users, contractors, and sensitive applications.
- Assign each AI agent its own identity and restrict its permitted destinations.
- Set a VPN retirement schedule rather than retaining indefinite fallback access.
The immediate defensive task is to establish whether a vulnerable gateway has already been used. The longer-term task is to reduce the internal reach that a future compromise could provide. Network leaders should connect those efforts: investigate access after each critical flaw, then use the findings to prioritize which connections move to application-level controls first.
