Latest / Dark Perimeter: Real Breaches, Real Stakes / Critical Is Not a Priority
Transcript
- 0:00SPEAKER_02: Four things carried a critical severity rating this week.
- 0:04SPEAKER_02: One of them had a federal remediation deadline that is already passed.
- 0:08SPEAKER_02: One of them has no finished patch for every supported branch.
- 0:13SPEAKER_02: One of them is a file red that only matters if the attacker already knows the file name.
- 0:19SPEAKER_02: And one of them was 247 fixes shipped in a single browser release.
- 0:25SPEAKER_02: All four are critical.
- 0:27SPEAKER_02: They are not remotely the same problem.
- 0:29SPEAKER_02: And if your program treated them the same way this week, your program spent its attention in roughly inverse proportion to your risk.
- 0:38SPEAKER_02: I'm Cole Draden.
- 0:40SPEAKER_02: This is Dark Perimeter.
- 0:42SPEAKER_02: With me are Dr.
- 0:43SPEAKER_02: Elliot Vance and Marcus Hale, and we are going to rank the week out loud with reasons.
- 0:49SPEAKER_02: Because the ranking is the actual work.
- 0:52SPEAKER_00: I want to set the terms first because the word critical is doing two different jobs, and that is the root of the confusion.
- 0:58SPEAKER_00: A CVSS base score describes the vulnerability and isolation.
- 1:03SPEAKER_00: Network reachable, no privileges required, no user interaction, high impact on confidentiality, integrity, and availability.
- 1:11SPEAKER_00: That is what produces a 9.8.
- 1:13SPEAKER_00: It is a description of the shape of the hole.
- 1:15SPEAKER_00: It is deliberately not a statement about whether anyone is standing at that hole with a shovel.
- 1:20SPEAKER_00: Whether you own the affected product, or whether that product sits on the internet, or on a management VLAN eleven people can reach.
- 1:27SPEAKER_00: Those are the facts that determine urgency, and none of them are in the score.
- 1:31SPEAKER_01: And in a real program, you have a finite number of change windows, a finite number of people who can take a management plane down, and a change advisory board that is going to ask you why this one and not that one.
- 1:41SPEAKER_01: So when four criticals land in a week, you are not choosing between patching and not patching.
- 1:46SPEAKER_01: You are choosing an order.
- 1:47SPEAKER_01: Everybody chooses an order.
- 1:48SPEAKER_01: The only question is whether the order was reasoned or whether it came out of whichever vendor emailed loudest.
- 1:53SPEAKER_02: Then let's reason it.
- 1:56SPEAKER_02: Top of my list this week is Cisco.
- 1:58SPEAKER_02: Vance Lay It Out.
- 2:00SPEAKER_00: Cisco Catalyst, SD WAN Manager, CVE 2026 76504, CVSS 9.8.
- 2:09SPEAKER_00: The mechanism is improper handling of URL encoding, classified as CWE 177.
- 2:15SPEAKER_00: There is an authentication rule intended to restrict access to a particular API endpoint, and a crafted request that percent encodes a character slips past the rule.
- 2:25SPEAKER_00: Cisco's own description is that encoding any single character in the request can be sufficient.
- 2:30SPEAKER_00: A successful request yields API access as the admin user.
- 2:34SPEAKER_00: Cisco published the advisory on September 30th.
- 2:37SPEAKER_00: SIESA added it to the known exploited vulnerabilities catalog the same day with a federal civilian remediation date of October 3rd, which is now passed, and the Kev entry also calls for a compromise assessment rather than only a patch.
- 2:51SPEAKER_01: The detail that moves this to the top for me is how Cisco found out.
- 2:55SPEAKER_01: Per the reporting, Cisco's incident responders learned of active exploitation in September through a tax support case.
- 3:01SPEAKER_01: That is not a researcher disclosure, and it is not a honeypot.
- 3:04SPEAKER_01: That is a customer calling support about something that had already happened to them.
- 3:07SPEAKER_01: There is a difference between a vulnerability that is theoretically exploitable and a vulnerability whose existence was discovered by cleaning up after it.
- 3:15SPEAKER_02: And the asset is the management plane for your wide area network.
- 3:19SPEAKER_01: Which is the part people keep underrating.
- 3:21SPEAKER_01: SD WAN manager is where the overlay policy lives.
- 3:24SPEAKER_01: If someone holds admin API on that, they are not in one branch office.
- 3:28SPEAKER_01: They are in the thing that decides how every branch office wrote.9 through 26.2 lines, and anything older than 20.9 has to migrate rather than patch.
- 3:40SPEAKER_01: Cisco says there is no workaround for on-premises deployments.
- 3:44SPEAKER_01: The guidance is the usual compensating set, restricted off the internet, limit to known hosts, put it behind a filter, and Rapid7's write-up is explicit that you apply the update anyway.
- 3:53SPEAKER_00: One more thing worth saying carefully because it is easy to blur.
- 3:56SPEAKER_00: This is reported as the fifth SD10 day Cisco has disclosed this year.
- 4:02SPEAKER_00: Earlier in the year, there was a separate pair of unauthenticated peering authentication flaws in the same product family CVE 2026-200127 and CVE 2026-2182.
- 4:14SPEAKER_00: Talos tracked exploitation of the first of those under the cluster name UAT 861 back in February and said it found evidence of related activity going back at least to 2023.
- 4:26SPEAKER_00: Different bug, different authentication path.
- 4:29SPEAKER_00: Nobody has publicly attributed this week's CVE 202676504 to that actor or to anyone else.
- 4:36SPEAKER_00: And Cisco has not shared details about the attacks.
- 4:39SPEAKER_00: The pattern across the year is real.
- 4:41SPEAKER_00: The attribution for this specific vulnerability does not exist yet.
- 4:45SPEAKER_00: And if you heard it stated as fact, that was inference wearing a suit.
- 4:49SPEAKER_02: Noted.
- 4:51SPEAKER_02: Second on my list, 40Net.
- 4:54SPEAKER_02: Hail, this is the one I expect to annoy people.
- 4:58SPEAKER_01: It should.
- 4:58SPEAKER_01: FortyMail, CVE 2026 1004286, also 9.8.
- 5:03SPEAKER_01: Path traversal combined with improper handling of null bytes, and the result is that an unauthenticated attacker writes arbitrary files on the appliance through a crafted HTTP or HTTPS request.
- 5:13SPEAKER_01: Fortinet credits its own product security team, Gwendol Gueno, with discovering and reporting it, and Fortinet has acknowledged exploitation in the wild.
- 5:21SPEAKER_01: SISA added it to Kev on October 1st with a federal deadline of October 4th.
- 5:25SPEAKER_02: So why is it second and not first?
- 5:28SPEAKER_01: Because of exposure arithmetic, not because it is less nasty.
- 5:31SPEAKER_01: Arbitrary file write as unauthenticated on a mail gateway is as bad as it reads, and the published indicators tell you what it is being used for.
- 5:38SPEAKER_01: The reporting lists added files including a shared object under the data library path, a web console binary, a mail service binary, and an L.so.preload entry, plus modifications to the HTT configuration and the SMIT binary.
- 5:51SPEAKER_01: An LD preload entry is not reconnaissance.
- 5:54SPEAKER_01: That is somebody installing a loader that runs inside every process that starts afterward.
- 5:59SPEAKER_01: If you have those files, you do not have a patching problem, you have an incident.
- 6:03SPEAKER_00: The reason I would still rank it just under Cisco is the denominator.
- 6:07SPEAKER_00: Every organization with an SD-WAN deployment has exactly one management plane, and it is central by design.
- 6:14SPEAKER_00: FortyMail is a product with a narrower install base than FortyGate, and the mitigating guidance hits the two conditions that make it reachable.
- 6:21SPEAKER_00: Disable the identity-based encryption support through the CLI, and restrict management interface access off the internet to trusted private networks.
- 6:30SPEAKER_00: If both of those were already true in your environment, your exposure to this is materially lower.
- 6:36SPEAKER_00: I would add the part that irritates me most though.
- 6:38SPEAKER_00: Several fixed builds were described in the advisory as upcoming rather than released.
- 6:45SPEAKER_02: Because that is a real operational condition, not a footnote.
- 6:50SPEAKER_00: It means that for some branches, the workaround is the control, not a bridge to the control.
- 6:54SPEAKER_00: People have been trained for 20 years to treat mitigations as a holding action until the patch, something you accept for a weekend.
- 7:01SPEAKER_00: If the patch does not exist for your branch, the configuration change is the remediation, and it has to be tracked and verified as such, with somebody owning the recheck for when the build ships.
- 7:11SPEAKER_00: The failure mode is not negligence.
- 7:13SPEAKER_00: It is a ticket closed as mitigated and never reopened.
- 7:17SPEAKER_01: Mitigated is not fixed, and a mitigation with no scheduled recheck is a decision to forget.
- 7:24SPEAKER_02: Atlassian.
- 7:37SPEAKER_00: CVSS 9.3.
- 7:39SPEAKER_00: Unauthenticated arbitrary file access across eight self-hosted data center products Bitbucket, Bamboo, Crowd, Crucible, Confluence, Fisheye, Jira Service Management, and Jira Software.
- 7:53SPEAKER_00: Atlassian's characterization is narrow and worth reading literally.
- 7:56SPEAKER_00: An unauthenticated attacker can reach specific files in the web application root directory.
- 8:02SPEAKER_00: Exploitation requires prior knowledge of the target file's exact name and path.
- 8:06SPEAKER_00: And the flaw does not permit directory listing.
- 8:09SPEAKER_00: Atlassian's own caveat is the sentence that matters.
- 8:12SPEAKER_00: In some configurations, there may be sensitive files present that increase your risk.
- 8:17SPEAKER_01: And Watchtower found the configuration that turns a constrained file read into a full compromise.
- 8:22SPEAKER_01: In crowd-integrated deployments, there is a properties file that holds credentials in plain text at a predictable path.
- 8:27SPEAKER_01: Their threat intelligence specialist, Jordan Ganshev, made the point that crowd-based single sign on setups are the exposed case, because the authentication material can be pulled from a path you can guess.
- 8:37SPEAKER_01: If your single sign on broker's credentials are readable, the file read is not the vulnerability, it is the delivery mechanism.
- 8:44SPEAKER_02: Now the dispute.
- 9:31SPEAKER_00: What they released was a checker script that tells you whether your instance is vulnerable.
- 9:36SPEAKER_01: Which matters because those are two different stories about the industry.
- 9:39SPEAKER_01: One is researchers shipped a weapon and attackers picked it up.
- 9:42SPEAKER_01: The other is researchers shipped an explanation and attackers rebuilt it themselves in an afternoon.
- 9:47SPEAKER_01: The second one is what appears to have happened, and the second one is the more useful lesson because you cannot fix it by asking researchers to be quieter.
- 9:54SPEAKER_02: And the ranking consequence?
- 9:56SPEAKER_01: Honeypot attempts are evidence of scanning and attempted exploitation.
- 10:00SPEAKER_01: They are not evidence that anybody's confluence got taken.
- 10:02SPEAKER_01: As of this recording, I have seen no confirmed compromise of a production instance reported, so it ranks third, but it is the one with the steepest slope.
- 10:10SPEAKER_01: It is also the only one of the four where I would tell someone to go look at access logs tonight rather than schedule a hunt because the attempts are already arriving, and the indicator is cheap.
- 10:18SPEAKER_01: Unexpected, unauthenticated requests reaching paths under the application route.
- 10:23SPEAKER_02: Fourth, and this is the one I think gets misranked most often in the other direction.
- 10:28SPEAKER_02: Chrome.
- 10:30SPEAKER_00: Google released Chrome 155, build 8059.39, with 247 security fixes.
- 10:38SPEAKER_00: Four are rated critical, and all four are use after free.
- 10:41SPEAKER_00: Google has not stated that any of them are being exploited in the wild.
- 10:44SPEAKER_00: Two of the four in the component tables are credited to Xinyang Gay at Anthropic, assisted by Claude, which is its own interesting data point about where bug discovery volume is heading.
- 10:56SPEAKER_01: 247 is a number that gets screenshotted and sent to executives, and it means close to the opposite of what the screenshot implies.
- 11:03SPEAKER_01: A browser with a mature fuzzing and bounty pipeline producing a very large fix count is a sign the pipeline is working.
- 11:09SPEAKER_01: The delivery mechanism is an auto update most of your fleet takes without a change window.
- 11:13SPEAKER_01: The operational task is verifying update compliance and restart rates, which you should be measuring anyway.
- 11:19SPEAKER_02: So the order is Cisco, Fortinet, Atlassian, Chrome.
- 11:23SPEAKER_02: And I want to be clear about why in one sentence each.
- 11:27SPEAKER_00: Cisco first, because exploitation is confirmed by the vendor.
- 11:31SPEAKER_00: The discovery path was a victim calling support.
- 11:33SPEAKER_00: The asset is a central management plane, and the federal deadline has already elapsed.
- 11:38SPEAKER_00: Fortinet second, because exploitation is confirmed, and the post-exploitation indicators show persistence tooling, but exposure is narrower and partly configuration dependent.
- 11:48SPEAKER_00: With the complication that for some branches, the mitigation is the remediation.
- 11:53SPEAKER_00: Atlassian third, because exploitation is attempted and not yet confirmed successful.
- 11:57SPEAKER_00: The primitive is constrained unless crowd makes it unconstrained and the slope is steep.
- 12:02SPEAKER_00: Chrome fourth, because volume is not urgency and the patch delivers itself.
- 12:06SPEAKER_01: And I would add a fifth item with no CVSS score at all that would outrank Chrome on any list I built.
- 12:12SPEAKER_01: The FBI and the Secret Service were reported this week to have warned that a credential theft campaign against Internet-facing Fortinet devices remains active, with a figure of more than 86,000 devices across 194 countries, and access sold on to ransomware affiliates.
- 12:27SPEAKER_01: Treat that device count as reported rather than confirmed.
- 12:29SPEAKER_01: I have it from aggregated coverage, not from the bulletin.
- 12:32SPEAKER_01: But the mechanism needs no vulnerability.
- 12:34SPEAKER_01: It is legacy hashes and reuse passwords on a box facing the internet, and there is nothing to patch.
- 12:40SPEAKER_02: Which brings us to the thing that actually happened to the most people this week, and it had no CVE in it anywhere.
- 12:48SPEAKER_02: Denmark.
- 12:49SPEAKER_00: Denmark's digitalization ministry announced on October 5th that unauthorized parties accessed names, addresses, and CPR numbers, the national personal identification number, for approximately 8.8 million people.
- 13:02SPEAKER_00: The register holds about 11 million records because it includes people who have died or emigrated, against a resident population of roughly 6 million.
- 13:11SPEAKER_00: So that is about four in five of everyone in the register.
- 13:14SPEAKER_00: The ministry's statement is that the access was obtained using a private Danish company's lawful right to look up records.
- 13:21SPEAKER_00: The company has not been named and is described as small.
- 13:24SPEAKER_00: There is no attribution.
- 13:26SPEAKER_00: Data Tilsonet, the Data Protection Authority, characterized the activity as a very large number of automated lookups made to identify valid CPR numbers and used the word allegedly about the numbers retrieved because its own assessment is not complete.
- 13:41SPEAKER_01: The timeline is the part I would read to a board.
- 13:43SPEAKER_01: Per the minister, Christina Egelin, speaking through Danish Wires, the access ran for about 10 days in September.
- 13:49SPEAKER_01: An employee of the Register Administration noticed the unusual activity on Friday, October 2nd.
- 13:54SPEAKER_01: Over that weekend, they worked out how many records were involved.
- 13:56SPEAKER_01: The Data Protection Authority was notified on the 4th.
- 13:59SPEAKER_01: The public announcement came on the 5th.
- 14:00SPEAKER_01: Egelin's own words, as reported, were that the safeguards were not solid enough, and that alarms should have sounded sooner, and she has asked for a full security review.
- 14:08SPEAKER_01: She also said it is too early to say whether people will need new CPR numbers, which is the sentence that tells you how bad this is, because a national identifier you may have to reissue is not a password.
- 14:19SPEAKER_02: There is a conflict in the record on one point, and I want it named rather than resolved.
- 14:25SPEAKER_00: There is on whether the company's access was cut off.
- 14:28SPEAKER_00: The Euronews account on October 5th, summarizing the government's statements, says the access had not been revoked.
- 14:34SPEAKER_00: The Hacker News account, citing the register administration, says the administration stopped the company's access with no date given, and other coverage says the access was terminated.
- 14:44SPEAKER_00: Those cannot both be true at the same moment.
- 14:46SPEAKER_00: The most likely explanation is that they describe different moments on a fast-moving day.
- 14:51SPEAKER_00: And the earliest statement was simply overtaken.
- 14:53SPEAKER_00: I am not going to pick one and present it as settled.
- 14:56SPEAKER_00: If you are briefing this internally, the honest line is that the access was reported revoked, with at least one same-day account saying otherwise, and that the discrepancy is unresolved in public reporting.
- 15:07SPEAKER_01: And then the question nobody has answered, which is the one I would actually want answered.
- 15:11SPEAKER_01: Register guidance limits a company's lookups to people it already has dealings with.
- 15:15SPEAKER_01: The Hacker News made this point, and it is the right one.
- 15:18SPEAKER_01: No official statement explains how one small company's legitimate access reached four and five people in the country.
- 15:24SPEAKER_01: Either the authorization model permitted far broader lookups than the guidance describes, or the rate and pattern of lookups was never bounded or both.
- 15:31SPEAKER_01: And whoever got into that company's systems, which is also unexplained, did not need a vulnerability in the register at all.
- 15:37SPEAKER_01: They needed a credential and permission that nobody had put a ceiling on.
- 15:41SPEAKER_02: So put the two halves of this episode together for me.
- 15:45SPEAKER_00: The four criticals are a cue management problem, and a queue has an order whether or not you choose it deliberately.
- 15:50SPEAKER_00: Score tells you the shape of the whole.
- 15:52SPEAKER_00: Exploitation status, exposure, asset centrality, and patch availability tell you when to act.
- 15:58SPEAKER_00: Those are different inputs, and only the first one arrives pre-calculated, which is exactly why it dominates programs that have not decided otherwise.
- 16:10SPEAKER_01: The largest event of the week by population affected involved no exploit, no severity score, no advisory, and no patch.
- 16:16SPEAKER_01: It involved standing access that somebody granted lawfully, a limit that was either absent or unenforced in ten days before a human being noticed.
- 16:23SPEAKER_01: There was nothing to patch, which means a program organized entirely around patching would have had no place to put it.
- 16:33SPEAKER_02: Cisco's SD WAN management plane, confirmed exploited, and already passed its federal deadline.
- 16:49SPEAKER_02: Atlassian, attempted and not yet confirmed.
- 16:53SPEAKER_02: Dangerous mostly where crowd keeps credentials in a guessable file.
- 16:58SPEAKER_02: Chrome, nothing to do but confirm your fleet took the update.
- 17:02SPEAKER_02: Then 8.8 million people in a country of six million from delegated access that nobody bounded.
- 17:10SPEAKER_02: If you're weak, went to the browser and not to the management plane.
- 17:14SPEAKER_02: The score picked your priorities for you.
- 17:17SPEAKER_02: This is dark perimeter.