Handed Claude Code our vendor list and said "which of these aren't marked trusted yet." It found 6 missing, suggested settings for each, asked to confirm, then just... did it. No clicking through 6 separate entries. Feels like cheating, doesn’t it? hubs.li/Q04ylbHg0

How safe is your Google Workspace? Take our free 5-minute assessment and share your results in the comments. 🔗 tinyurl.com/25cv6z5y

A malicious invite looks like it's from a recruiter, a vendor, your boss, which is exactly why attackers moved there. The email carrying it is clean, no link, no attachment, because the payload sits in the event's Location or Description field. Both Google Workspace and M365 auto-add that event to the calendar by default, no click needed, so your filters never see it. How to fix it: → Google Workspace: Admin Console → Calendar → Advanced settings → only add invites the user responds to → Microsoft 365: per mailbox via PowerShell (Set-CalendarProcessing -AutomateProcessing None) → Tell your team what these look like, since users can override both settings → Have a cleanup plan for events that already landed, compromised accounts still get through

⏰ Happening tomorrow See how real Google Workspace breaches used social engineering and malicious OAuth apps—and what happened during the critical first hours of response. Join our Sept. 23 webinar with @material_sec. 🎯 bleepingcomputer.com/rd/102/ #cybersecurity #GoogleWorkspace

1,064 apps in our data have more historical authorizations than currently authorized accounts. That's the same gap Apple's July lawsuit describes, a former employee who kept internal access for months because the grant, not the account, never got revoked. Full findings in our Workspace Risk Index Vol. 1: hubs.li/Q04y1bH90

The 90s toolbar promised convenience, delivered malware. Same trick today just wears a "Sign in with Google" button instead of a browser icon. Nosferatoolbar: you should have read the permissions screen.

Password reset = fresh start, right? Wrong. OAuth tokens don't care. They're minted per app and survive a reset completely untouched, still valid, still working, still yours (well, theirs now). Offboarding kills them. Nothing else does. Tie your OAuth grants to a personal account and they're stuck living and dying with that human. Tie them to a service account and they don't vanish when someone changes their password, quits, or gets hit by a bus (dark, but true). Use a service account. Your IT sec team is not going to stop saying it.

An employee connects an AI tool to Google Workspace in one click. No ticket, no review, no separate permission tier for what the AI can reach versus what the employee can. It just inherits everything. Across 159 orgs, we found 978 of these connections doing exactly that. Full report: hubs.li/Q04xPg3t0

"Is it normal to feel a little betrayed by your own IT team after a fake phishing test?" @Quora's own IT team clearly thought so, they ditched the quiz. Now one real phishing report warns everyone else instantly, no test, no shame video. How 5 people pulled it off: tinyurl.com/yymeekrk

Whale phishing is illegal no matter how you look at it, right? Protecting our whales matters. In this case, the whales are your CEO and CFO, the people whose approval alone can move money. Levitas Capital lost $8.7M to a fake Zoom invite. Arup lost $25M on a deepfake video call. One habit stops most of it: verify financial requests through a second channel, every time. hubs.li/Q04xBjKN0

Rajan built a malicious OAuth app, start to finish, for about $5 in cloud costs. Then he showed Matt Johansen how it steals a mailbox without ever tripping an account takeover alert. No MFA bypass. No new login. Just a token doing what it was scoped to do. Full watch: hubs.li/Q04xBhVZ0

Why do attackers want into your cloud workspace instead of just your password? Because OAuth grants survive what passwords don't. A reset doesn't revoke it, MFA doesn't stop it, and it doesn't trigger a login alert. Rajan Kapoor gets into this in a clip from our OAuth webinar, including almost testing it on our very own CEO.

Two ways to get someone to trust you without earning it. Angler phishing is a footrace. You complain publicly about a company, and a fake support account beats the real one to the reply, same logo, same apologetic tone, just enough resemblance to pass a quick glance. It asks you to move to DMs. You do, because you're already annoyed and just want the problem solved. None of this touches email, so DMARC never sees it happen. Clone phishing is patience. It takes an email you already opened once, something you trusted, copies it almost exactly, and swaps the link. If the attacker replays a genuinely signed message instead of forging one, it can pass DKIM too. It's not trying to look official. It's trying to look like something you've already filed away as safe. Different clocks, same bet: that you'll skip the second look because you already gave the first one. Clone phishing: tinyurl.com/5x34s6t7 Angler phishing: tinyurl.com/mrkbftb7

Every organization we studied has sensitive data sitting in email. Not most. All of them. We analyzed 159 Google Workspace environments and found 373 million sensitive emails, 475 million sensitive Drive files, and 978 active OAuth apps most IT teams never approved. AI agents are already reading that data. Full findings in our Workspace Risk Index Vol. 1: hubs.li/Q04x30bx0

Our Staff Threat Research Engineer Belem Regalado just wrapped an OAuth governance investigation across 159 organizations. Three fixes came out of it, and none need a new tool. 1. Add a line to offboarding. ClockWise shut down 2,303 accounts across 21 environments. 60 days later, every account still held active OAuth grants. Offboarding caught zero, because disabling an account and revoking what it authorized are two different steps. 2. Capture ownership at approval time. Who approved it, what it's for, when to review it again. Three fields. A spreadsheet works. 3. Actually review on that cadence, so six months from now someone can answer why an app still has access. Small habits, rather than new software.

VPNs killed. ZTNA deployed. MFA everywhere. Then a user logs into Google Workspace and the mindset flips from assume breach to assume safety. A compromised mailbox isn't today's email, it's a decade of M&A docs and HR discussions sitting wide open in the archive. hubs.li/Q04wSKs_0

Mythos had three weeks of hype. It left one useful question: what happens after detection fails? Email's been living that answer since 2016. There's no patch cycle for a stolen password. hubs.ly/Q04wLMrQ0

Your healthcare BA just got breached! You have 60 days to report it to HHS, even though you didn't cause it and their timeline isn't yours. What to do first: - Confirm which mailboxes were actually compromised. Don't accept 'we're looking into it.' - Check your BAA for notification and cooperation terms before you need them. - Assume the exposure is bigger than reported. Vendors often disclose what was accessed, not everything sitting in the mailbox. - Get legal involved immediately. This is your incident too. Why it happens: BAs cause 15% of healthcare's email breaches but 42% of exposed records. One compromised vendor mailbox can hold PHI from every client they serve.

An AI agent doesn't need to be compromised to walk the same path as an attacker. Ambiguous instructions or one stray prompt in its environment is enough for it to read inbox content beyond its task, act on credentials and documents it finds, send messages, and ask for access to the next system. Nothing malicious, nothing stolen. Just an agent doing its job inside guardrails that were never built for it. More from Rajan Kapoor: hubs.li/Q04wrkL50

The ChatGPT Google Workspace connector holds 21 OAuth scopes. Full Gmail read/write. Every calendar event and attendee. Your entire contact graph. Every employee in your org directory. Every file in Drive, not just recent ones. One click grants all of it. More from Belem Regalado👇