Material Security

Protect your critical infrastructure. Material Security is a multi-layered detection and response toolkit for Google Workspace and Microsoft 365

[Material Security @material_sec]

@DevCalledFede's exact problem (a session that only exists locally) is why scoped, revocable grants matter more than sandboxes. OAuth already has that concept, most local dev credentials don't.

Same tension as @thepanta82's point, just showing up elsewhere: broad standing access is more useful and also the thing nobody reviews six months later. Scoped-but-revocable beats broad-but-convenient whether it's a coding agent or one connected to Gmail and Drive.

[Material Security @material_sec]

The "agent takes on the role of the user" vs. "agent gets its own identity" split you mention is probably the highest-leverage decision in that whole list, more than model choice or orchestration layer.

An agent acting as the user inherits whatever that user can already touch, no new access review triggers because nothing new was technically granted.

An agent with its own identity at least creates a place to scope, log, and revoke access deliberately.

Most of the "substantial guardrails vs. responsibility on the user" split you mention downstream is really just a consequence of which of those two paths got picked upstream, once an agent is wearing a user's identity, guardrails become a policy problem instead of an access-control one.

[Help Net Security @helpnetsecurity]

Product showcase: Material unifies Google Workspace email, file & OAuth defense - https://t.co/PrBH9Xj8Tw - @material_sec #GoogleWorkspace #EmailSecurity #Cybersecurity #CybersecurityNews

[Material Security @material_sec]

It's go time. #BlackHat booth #6021, "Stop Securing Email."

The team's on the ground: Boomer, Jordan, Ahmad, Chandler, Austin, Alex, Michael.

Come talk OAuth risk and AI agents. Grab the new Risk Report while you're there!

https://t.co/KZB6CEUXOD

[Material Security @material_sec]

The overlooked variable in 'AI SOC investigates the full queue' is what access it needed to do that. An agent that can pull data across email, cloud storage, and endpoints to hunt effectively is holding a pretty wide set of permissions. Worth watching that as its own thing, separate from whether the hunting itself is any good.

[Material Security @material_sec]

We taught our debug agent the one thing even engineering interns take years to learn: guess first, then go check: https://t.co/tsS8b2QQQR

[Material Security @material_sec]

@tqbf @DennisF @matthew_d_green Okay fair, $100k isn't a big number in this space but the timeline is still worth watching - a specific structural weakness sitting in a scheme that had two years of dedicated review.

[Material Security @material_sec]

"Stop Securing Email."

Email security is foundational. Yet, it is not enough.

- 13x rise in AI phishing
- 25% of OAuth apps one bad click from disaster
- 45% of orgs with zero OAuth governance
- 1,100% growth in sensitive files nobody's watching

Come to booth # 6021 at #BlackHat. We'll explain exactly what we mean: https://t.co/ROlgstTK8Z

[Material Security @material_sec]

Your email gateway is still fighting 2016's battle.

The real threat today is someone who read your LinkedIn, knows your CFO's name, and asks AP to wire money. Nothing to scan, nothing to flag, it just looks like Tuesday.

@Gerald_Auger spent 10 days testing Material Security against exactly that. It caught what the gateway would've waved through.

[Material Security @material_sec]

Somewhere in your Drive: a doc called "final_final_v2_ACTUAL_final," shared with "anyone with the link." Who knows what’s in it?

New in Material:
→ Dashboard mapping sensitive file sharing across Drive
→ Plain-English file search
→ One-screen custom detections

Less sprawl, more signal: https://t.co/FrOYtS6cvr

[Material Security @material_sec]

Congrats on shipping this - the isolation-at-runtime problem is one nobody's solved well yet. Curious how you think about the layer above this: an agent can be fully sandboxed and still legitimately read something sensitive it had access to (that's the gap we spend most of our time on). Excited to dig into this!

[Material Security @material_sec]

The build-time side of this is maturing fast, OWASP checks on agent-generated code are becoming standard. The gap is still upstream: OAuth grants tell you what an agent is allowed to do, not what it's actually doing, and that mismatch is where most of the risk sits once an agent is live. We looked at 22K+ OAuth apps recently and found a lot of scope that nobody was watching in practice.

[Material Security @material_sec]

Our booth at Black Hat says "Stop Securing Email."

Yes, we know how that sounds coming from an email security company. Bear with us.

Cut off one phishing attempt and OAuth, session tokens, and rogue AI agents grow back as five more heads. Email was never the whole beast.

Booth [# 6021], Aug 4-6. Come argue with us about it.
🖇️ https://t.co/EeyIPnoySS

#BlackHat @BlackHatEvents

[Material Security @material_sec]

@tqbf @DennisF @matthew_d_green Mainly the cost curve. Two years of NIST review is close to as much expert scrutiny as a candidate algorithm gets before deployment, and this took ~60hrs / ~$100k to find something that review missed.

[Material Security @material_sec]

@frgx 'We held the tool back... until we could look them in the eye and say this is worth your time.' That sentence should be printed out and taped above every agent security roadmap.

[Material Security @material_sec]

Your security stack has more silos than a Midwest cornfield. Material's MCP server lets your AI agents actually talk to it, investigate, act, remediate, no console-hopping required:
https://t.co/HAeVHFKyXN

[Material Security @material_sec]

Rajan Kapoor, Material's VP of Security, built a demo app just to show how fast this actually goes.

Phishing email → click "Allow" → mailbox fully exfiltrated → live view into Drive → pivoting to Slack. Total time: under a minute. No password. No MFA prompt.

His words on building it: "I just woke up and chose violence one morning."

Here's why it works: OAuth isn't authentication, it's authorization. It's built so access survives a password reset, on purpose, so a connected app doesn't break every time you update a password. Attackers exploit that same design: reset the password, revoke nothing, the token keeps running. Offboarding barely touches it either.

Full breakdown on why consent is the new vulnerability:

https://t.co/IUycfe3W1e

[Material Security @material_sec]

Here’s @Gerald_Auger 's take on #MaterialSecurity catching a malicious email in Google Workspace.