Formal Intel

Intelligence from the frontlines of AI and cybersecurity. By @joinformal.

[Formal Intel @FormalIntel]

Introducing Formal Intel.

Cybersecurity news, emerging threats, and analysis of how AI is changing security.

We're powered by @joinformal, which enforces fine-grained access controls at the network layer so AI agents can scale with sensitive systems without excessive permissions.

Least privilege for a world of agents.

[Formal Intel @FormalIntel]

@jfrog The interesting question is how government teams can make supply chain security continuous rather than a compliance checkpoint. As AI accelerates development, can existing review and approval processes keep pace?

[Formal Intel @FormalIntel]

@OpenAI This raises an interesting question for security teams: as AI dramatically increases vulnerability discovery capacity, will patch review and deployment become the new bottlenecks?

[Formal Intel @FormalIntel]

Least privilege for agents is the right idea, but there are still cases that can allow an agent to modify something you wouldn't have expected (like a build script) to overreach its intended privileges. The missing piece most teams still struggle with is just-in-time access.
Temporary, narrowly scoped permissions that expire automatically instead of keys that live with the agent.

[Formal Intel @FormalIntel]

@_shreya_s You've mirrored my thoughts on it exactly. The ability for AI to easily find and exploit vulnerabilities in code means that vendors really have to go over their product with a fine-toothed comb. We are only now coming to grips with some of the methodologies to do this.

[Formal Intel @FormalIntel]

@BlackHatEvents The company owns the risk. The question is whether they can prove *which* human authorized the agent's actions, and what it was actually allowed to touch. This necessitates a granular framework for just-in-time authorization by a real human being for access to sensitive assets.

[Formal Intel @FormalIntel]

OIDC for CI/CD was progress. OIDC for cloud agents that can request scoped, time-bound prod access (and automatically expire) is the next step. Secrets being left accessible to runners and agent sandboxes is a risk we no longer need to accept.

[Formal Intel @FormalIntel]

Following CISA's active exploited list used to be good enough to know when to patch something. Most orgs now directly track cves only from https://t.co/etWu6nLRwd master list. Did you know that's too slow now? Most vendors publish first to their own PSIRT, this is where attackers are looking for new exploits. Your org shouldn't be waiting for PSIRT reports to show up on CVE master list. You should be aggregating these sources into your own list to ensure you can patch things as fast as possible.

[Formal Intel @FormalIntel]

@kimmonismus They're still subsidizing subscription based usage by like 90%... please don't ruin this for the rest of us :o

[Formal Intel @FormalIntel]

@UK_Daniel_Card Modern attackers take exactly ten minutes after a patch drops to make a weaponized exploit. Security through obscurity is kinda overkill, running your services on non default ports is probably the easiest way to avoid being on a mass exploit list.

[Formal Intel @FormalIntel]

@thedawgyg This is a really cool drop, I hope Github doesn't ban your account for it. Looking forward to playing around with these fuzzers today.

[Formal Intel @FormalIntel]

@IntCyberDigest Hands down one of the funniest priv esc methods I've seen since shellshock (I'm getting old, that was 12 years ago!)

[Formal Intel @FormalIntel]

@thsottiaux I'm basically the only guy looking forward to 6.2 Luna, it does pretty great with security research at a really low budget.

[Formal Intel @FormalIntel]

@ZackKorman OpenAI kinda cursed the whole world. You can do an offensive audit on any site now, all you gotta do is make it seem as if it was unintentional by your agent.

[Formal Intel @FormalIntel]

@IntCyberDigest This is going to become more common over time. The best way to protect your family from this is to create a 'safeword' for your family. My daughters know a special word they need to say on the phone for emergency situations.

[Formal Intel @FormalIntel]

@seanhn Defenders know their infrastructure better than attackers. This is still true in today's agentic world. Tripwire setups and proper permission scoping can still help people stay secure.

[Formal Intel @FormalIntel]

@clairejyz @FortuneMagazine @CylakeAI Formal (@joinformal) will be up next!

[Formal Intel @FormalIntel]

@joinformal Scaling AI agents? Your security needs to scale just as fast.

Book a demo: https://t.co/ZuYFf4xFh9

[Formal Intel @FormalIntel]

Introducing Formal Intel.

Cybersecurity news, emerging threats, and analysis of how AI is changing security.

We're powered by @joinformal, which enforces fine-grained access controls at the network layer so AI agents can scale with sensitive systems without excessive permissions.

Least privilege for a world of agents.

[Formal @joinFormal]

Meet @FormalIntel: our new home for cybersecurity news, emerging threats, and analysis of AI security.

Follow along:

[Formal Intel @FormalIntel]

@rauchg Impressive numbers! When AI agents start shipping security fixes autonomously, the security boundary shifts to what those agents can access and change.

Least privilege, scoped credentials, and audit trails become even more critical at this scale.