buc.ci is a Fediverse instance that uses the ActivityPub protocol. In other words, users at this host can communicate with people that use software like Mastodon, Pleroma, Friendica, etc. all around the world.
This server runs the snac software and there is no automatic sign-up process.
Microsoft demonstrated how AI coding agent device code phishing and OAuth device code phishing can hijack GitHub Copilot CLI sessions and private repos.
#DeviceCodePhishing #GitHubCopilot #AICodingAgents #MicrosoftSecurity #OAuth #Storm2372 #Cybersecurity
Proofpoint uncovered the BluePhishing Entra Agent attack, where one OAuth consent lets hackers spawn 250 rogue Microsoft Entra Agent ID accounts.
#BluePhishing #MicrosoftEntra #Cybersecurity #OAuth #AIAgents #CloudSecurity
CVE-2026-118669 – HIGH severity in python-oauthlib affects RHEL 9.8 EUS+; allows privilege escalation via client identity confusion in OAuth token grants. Update to 3.1.1-5.el9_8.1 ASAP. https://radar.offseq.com/threat/red-hat-security-advisory-python-oauthlib-security-update-cb303aa14cf05295 #OffSeq #RedHat #Vuln #OAuth #Linux
----------------
🎯 Threat Intelligence: BluePhishing and Entra Agent Identities
===================
Proofpoint describes BluePhishing, a research proof of concept that extends OAuth consent phishing to Microsoft Entra Agent ID. The company has not identified activity using the technique in the wild.
🔹 Model and permissions
Entra Agent ID uses an agent blueprint that corresponds to an application registration. A blueprint principal corresponds to a service principal, while an agent identity is an instance created from a blueprint. Microsoft documents a limit of 250 agent identities per non-Microsoft-managed blueprint in a tenant. By comparison, an application registration maps to one service principal per tenant.
The attacker prepares a blueprint in a tenant they control, sets its sign-in audience to AzureADMultipleOrgs, adds a client secret and configures permissions through requiredResourceAccess and inheritable permissions. The source says the audience cannot be changed in the Entra portal. Microsoft Graph, PowerShell or the manifest must be used instead.
🔹 Attack Chain Analysis
🎣 Initial access: The attacker sends a consent link, such as an adminconsent URL using the .default scope. The scope requests the permissions preconfigured in the blueprint.
⚙️ Principal creation: If an authorized victim grants consent, a blueprint principal appears in the victim tenant. Proofpoint says it receives AgentIdentity.CreateAsManager and that this permission cannot be revoked from the principal.
🔑 Token acquisition: The attacker uses their blueprint application ID and secret to obtain a client-credentials token from the victim tenant's token endpoint.
🤖 Identity creation: The token is presented to Microsoft Graph to create agent identities. Each creation requires a sponsor, specified as a victim-tenant user's object ID.
📊 Graph access: The attacker uses the autonomous agent flow and fmi_path to obtain Graph tokens for the identities. Proofpoint tested application permissions including User.Read.All and Group.Read.All, using GraphRunner to issue Graph requests.
🔹 Consent paths and artifacts
Admin consent is the broader path described in the source because application permissions apply across the tenant. The recipient must be able to grant that consent. User consent produces delegated permissions, which cannot be used autonomously in the described flow and require further consent through the on-behalf-of flow. Proofpoint observed silent inheritance only for delegated permissions granted through user consent.
The described flow also requires the victim tenant ID and a sponsor user. The source notes that the OpenID Connect configuration endpoint returns the tenant ID and that GetCredentialType can enumerate users.
🔹 Scope and limitations
The supplied article does not provide detection rules, indicators of compromise or mitigation guidance, and its final discussion of inherited permissions is incomplete. It cites Message Center post MC1506733 as saying default Entra users will be able to create agent blueprints from November 2026. That is a separate exposure consideration, not evidence of an observed campaign.
Reference: Proofpoint, “BluePhishing: Initial Access via Third-Party Agent Blueprint Consent.”
🔹 BluePhishing #MicrosoftEntra #OAuth #ThreatIntelligence
Consent phishing just moved up a level. Proofpoint documented a campaign this week they call BluePhishing.
Instead of a user granting an app permissions, the lure asks them to approve a third-party AI agent blueprint. Same OAuth consent machinery, new wrapper.
There's no fake page to spot. The consent screen is the real one, so 'check the URL' doesn't apply. What decides it is whether your policy and people stop the grant.
How are you handling third-party agent consent right now?
Our OAuth consent grant scenario tests exactly that, with training tied to whoever approved. More at https://phishu.net/oauth-consent-grant-phishing-simulation.html
More like this every Monday in our newsletter. https://phishu.net/newsletter
#OAuth #ConsentPhishing #IdentitySecurity #Microsoft365 #SecurityAwareness
@firstyear Your minor rant about slinging Oauth id tokens around inspired me to spend a few hours more with miserable Entra figuring out how to make the access tokens do what I want so - thanks, I guess?
Proofpoint researchers have identified a potential attack technique that abuses AI agent identities. ⚠️
In a new blog post, they show how attackers could use BluePhishing to get victims to consent to a malicious blueprint that lets an attacker establish agent identities and operate with granted privileges.
What the research found:
• BluePhishing could be the next evolution of #OAuth consent abuse. It extends traditional OAuth phishing to AI agents and abuses consent mechanisms.
• The challenges of governance around agent identities and consent flows increase the risk of scalable and automated post-exploitation.
While #BluePhishing is currently a proof of concept, the findings highlight how AI adoption could introduce new opportunities for attackers to abuse Entra ID blueprints.
Google One Tap with Auth0: a native mobile app gets a built-in, Action-free token exchange straight from the platform. A web app can't use that path (it's gated to native clients), so Custom Token Exchange does the same job with a validator Action you control.
Also worth knowing: Universal Login's own native FedCM toggle, especially if your tenant's Google connection already carries Token Vault.
Apache Kafka fixes 12 vulnerabilities, including CVE-2026-110985, which skips JWT signature verification and grants full cluster access. Patch now.
#ApacheKafka #Kafka #CVE2026110985 #JWT #OAuth #DataStreaming #DoS #PatchNow
OpenMage fixes 34 OpenMage vulnerabilities in Magento LTS, including a critical CSRF flaw that hands out API tokens. Upgrade to 20.19.0 or 20.18.2.
#OpenMage #Magento #MagentoLTS #CSRF #OAuth #eCommerce #PHP #CyberSecurity
Exposed Master Key Sparks LiteLLM Supply Chain Attack
A single misstep on May 27 - committing a master key to a public repository - ultimately led to a supply chain compromise on October 10, allowing a threat actor to extract a sensitive GitHub CLI OAuth token. The lapse occurred when the exposed key wasn't revoked quickly enough, leaving the door open to malicious access.
Reset the password, and the attacker still has the mailbox.
A Spiceworks piece this week spelled out why. With OAuth consent phishing, a reset never revokes the app the user approved, so the token keeps reading mail after the password changes.
Check it yourself today. Open the connected-apps page in your Microsoft or Google account and remove anything you don't recognize.
It is exactly what we run against authorized targets in our OAuth consent grant campaigns, then train whoever approved the app.
See how it works at https://phishu.net/oauth-consent-grant-phishing-simulation.html
We round up the week's phishing news in The PhishU Weekly if you want it in your inbox. https://phishu.net/newsletter
#OAuth #ConsentPhishing #IdentitySecurity #Microsoft365 #CyberAware #SecurityAwareness #CybersecurityAwareness #AwarenessMonth
One stolen ConnectWise Cloud Backup credential let attackers sign into 173 Microsoft 365 tenants, no password, no MFA, as The Hacker News reported this week. That app already held standing Graph access across every customer tenant.
Teach your team this one this month. A password reset and MFA do nothing to an OAuth app grant, so review the apps connected to your Microsoft 365 or Google account and revoke what you don't recognize.
How are you tracking which apps hold standing access right now?
See how we run the OAuth consent technique against authorized targets to show that reach first. https://phishu.net/oauth-consent-grant-phishing-simulation.html
More like this every Monday in The PhishU Weekly. https://phishu.net/newsletter
#OAuth #ConsentPhishing #IdentitySecurity #Microsoft365 #CyberAware #SecurityAwareness #CybersecurityAwareness #AwarenessMonth
https://thehackernews.com/2026/10/hackers-use-compromised-connectwise.html
CVE-2026-111823 (CRITICAL, CVSS 9.8) in Mage mage-ai <=0.9.80: OAuth session creation flaw enables unauthorized access if provider is unknown. No confirmed patch — check vendor updates. Details: https://radar.offseq.com/threat/cve-2026-111823-cwe-390-detection-of-error-condition-without-action-in-mage-mage-ai-d243a1403d673a6a #OffSeq #CVE2026111823 #MageAI #OAuth #Infosec
So, you may have seen some hyperbolic news stories saying that #infosec whitehats used #Claude to hack #OpenAI, making it sound like frontier labs were out here declaring a binary battle of the behemoths. #AI vs AI; a battle to the last agent's dying bit.
That's not what happened. I've just read the report from the guys at #Hacktron, and if I've read it correctly, the real attack surface here was #OAuth. If this proof of concept can be replicated in other environments, OAuth may be a major security issue going forward.
I dunno, maybe I misread. Here's the report.
My live Auth0 session and token demo grew three new patterns since I first wrote this up: CIBA against a real Guardian push (now with a Rich Authorization Requests toggle), the Device Authorization Grant, and a comparison against a real Hono app on Cloudflare Workers. The session view also moved out of a single profile page into a sidebar that's visible on every page in the app.
#tech #dev #security #cybersecurity #InfoSec #Vercel #breach #OAuth #AI