Cloud Shell flaw rated 8.7 CVSS, user action needed
Cloud Shell CVE-2026-85267 ·
In this episode
The facts
- CVE-2026-85267 is a flaw in Google Cloud Cloud Shell.
- CVE-2026-85267 is classed as CWE-1385 in MITRE's weakness list.
- According to its CVSS rating, CVE-2026-85267 can be exploited remotely over the network, without any credentials and only if a user takes an action. A successful attack would fully compromise the confidentiality, integrity and availability of the affected system.
- CVE-2026-85267 is rated 8.7 out of 10 on the CVSS severity scale.
Sources
The panel
Jordan Pike AI host
Here's what's new. CVE-2026-85267 is a flaw in Google Cloud Cloud Shell, and it's classed as CWE-1385 in MITRE's weakness list.
Jordan Pike AI host
According to its CVSS rating, CVE-2026-85267 can be exploited remotely over the network, without any credentials and only if a user takes an action. A successful attack would fully compromise the confidentiality, integrity and availability of the affected system. It's rated 8.7 out of 10 on the CVSS severity scale.
Jordan Pike AI host
Who should care: anyone whose people use Cloud Shell. Whether a fix exists isn't in what we have, so keep that in mind. Maya, what kind of bug are we dealing with?
Maya Okafor AI analyst
The facts give us CWE-1385, but not a plain-language description, so I won’t guess at the mechanism or code location. What the CVSS rating does tell us is the shape: network reachable, no credentials, but a user has to do something. That's the profile of a flaw where the attacker needs no account, only a person's cooperation. A complete fix closes the whole class, not one path to it. When a fix lands, check that, and look at the sibling code nearby.
Dex Moreno AI analyst
From the attacker's chair, 'user takes an action' is the one speed bump, and in general it's a low one. Reach the person remotely, get them to engage, and the rating says full loss of confidentiality, integrity and availability. A hosted shell is the kind of place people work from with access to other things, so the real question is what that session can reach downstream: other projects, stored secrets, anything it's trusted by. That's if the rating holds up.
Dex Moreno AI analyst
I love that the attacker's only requirement is a willing participant. Very collaborative. Truly, a team-building exercise. The target deserved better.
Jordan Pike AI host
Great culture fit, I'm sure.
Maya Okafor AI analyst
Here's where I push back on the usual 'it's user-interaction, so it can wait' instinct. That's a bridge, not a fix. Awareness training and tighter access are fine as a stopgap, but give them an end date and apply the vendor fix as soon as one ships. A compensating control that never expires turns into the plan.
Dex Moreno AI analyst
I don't disagree about the fix, but I'd order the work by exposure. Which of your people actually use Cloud Shell, and what can their sessions reach? Cut that blast radius first, because that works today and a fix may not exist yet. Severity is a number, exposure is your actual Monday.
Maya Okafor AI analyst
Fine, but exposure work is the bridge. The fix is the destination. We can agree on both as long as nobody forgets the second one.
Sam Kowalski AI analyst
From the forensics side: assume anyone could have been reached earlier than the disclosure date. The facts don't say anything about exploitation, so I won't invent indicators. What I will ask is how long your cloud audit logs are kept, because a user-action flaw leaves its trail in what people did and when. If the retention window is short, you can't answer the question at all.
Sam Kowalski AI analyst
Your audit logs go back forever, I'm sure. Right up to the meeting where someone set the retention to thirty days.
Jordan Pike AI host
That meeting had snacks, and no minutes.
Rina Takahashi AI analyst
Threat intel's job here is mostly to say what we don't have. The facts don't mention any group attribution or exploitation, so anyone who tells you who's behind this is making it up. Watch for that as the story spreads. If a claim shows up, ask who is making it and how confident they are.
Rina Takahashi AI analyst
I'm thrilled to report this flaw currently has zero threat actor names attached. Give the industry a week.
Jordan Pike AI host
The naming committee has been notified.
Jordan Pike AI host
So the panel agrees on the destination and argues about the order. Each of you, one action. Maya?
Maya Okafor AI analyst
Apply the fix as soon as one ships, then check that it closes the whole weakness class and not a single path.
Dex Moreno AI analyst
Work out who uses Cloud Shell and what their sessions can reach, then shrink that today.
Sam Kowalski AI analyst
Find out how long your cloud audit logs go back, and preserve them before you need them.
Rina Takahashi AI analyst
Treat any attribution you read as a claim, and ask who's making it.
Jordan Pike AI host
Monday, then: know who uses Cloud Shell, limit what their sessions can reach, confirm your log retention, and apply the fix once one is available.