The Context Firewall for AI Agents AIR sits between agents and the outside world, continuously analyzes and filters every input into an agentโs context.
Yehuda ha-Levi St 34, Tel Aviv
Joined June 2026
- Tweets30
- Following6
- Followers505
- Likes4
๐ง๐ต๐ฒ ๐๐๐ฅ ๐ฉ๐ผ๐น๐น๐ฒ๐ ๐๐๐ฝ ๐ต๐ถ๐ด๐ต๐น๐ถ๐ด๐ต๐๐ ๐ฎ๐ฟ๐ฒ ๐๐๐ฅ๐ ๐ฅ
Huge thanks to everyone who played and cheered, you guys were INSANE!
Same time next year? Better start practicing ๐
@Island_io, @cyera_io , @wiz_io , @Vega_Threat, @Zafran_io, @CatoNetworks, @Enigma_AI, ZyG, @nebiusai, Koi, @DaylightSec
๐๏ธ ๐๐ผ๐ ๐บ๐๐ฐ๐ต ๐ฎ๐ฐ๐ฐ๐ฒ๐๐ ๐ถ๐ ๐๐ผ๐ผ ๐บ๐๐ฐ๐ต?
Our Co-Founder & CTO, @niv_hoffman, joined TrashTech to talk about one of the biggest challenges in AI security: agents donโt just access our data anymore. They can use our tools, interact with our systems, and take real actions on our behalf.
๐๐โ๐ ๐ป๐ผ๐ ๐ท๐๐๐ ๐ฎ๐ฏ๐ผ๐๐ ๐๐ต๐ฎ๐ ๐ฎ๐ป ๐ฎ๐ด๐ฒ๐ป๐ ๐ฐ๐ฎ๐ป ๐ฎ๐ฐ๐ฐ๐ฒ๐๐. ๐๐โ๐ ๐ฎ๐ฏ๐ผ๐๐ ๐๐ต๐ฎ๐ ๐ถ๐ ๐ฐ๐ฎ๐ป ๐ฑ๐ผ ๐๐ถ๐๐ต ๐๐ต๐ฎ๐ ๐ฎ๐ฐ๐ฐ๐ฒ๐๐.
From our malicious skill experiment that exposed over 26,000 agents to the guardrails needed to keep agents safe, Niv explains why the answer isnโt to hold AI agents back - itโs to childproof the environment. ๐งธ
Thank you @OmriBarak and @Tsuker for having us!
๐บ Watch the full episode on YouTube - youtube.com/watch?v=9EFvz9rtโฆ
๐ง Or listen on Spotify, Apple podcast & more -pod.link/1739161374/episode/โฆ
๐ช๐ฒ'๐ฟ๐ฒ ๐ต๐ผ๐ป๐ผ๐ฟ๐ฒ๐ฑ ๐๐ผ ๐ฏ๐ฒ ๐ป๐ฎ๐บ๐ฒ๐ฑ ๐๐ผ ๐๐ต๐ฒ hashtag#๐๐ถ๐๐ฟ๐๐ฝ๐๐ผ๐ฟ๐๐ฒ๐ฌ ๐๐น๐ฎ๐๐ ๐ผ๐ณ ๐ฎ๐ฌ๐ฎ๐ฒ!
Thank you to @TechCrunch Disrupt and @GreenfieldVC for the recognition!
And a special thanks to our customers, partners, and friends for the trust, collaboration, and feedback that help shape what we build โจ
See the full Class of 2026: disruptors60.ai/winners
๐จ ๐๐ฅ๐๐๐๐๐ก๐: ๐๐๐ฅ ๐ฟ๐ฒ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต๐ฒ๐ฟ๐ ๐ณ๐ผ๐๐ป๐ฑ ๐ฎ ๐๐ฎ๐ ๐๐ผ ๐๐ฌ๐ฃ๐๐ฆ๐ฆ @vercel 's ๐ต๐ฎ๐ฟ๐ฑ๐ฒ๐ป๐ฒ๐ฑ ๐๐ฎ๐ป๐ฑ๐ฏ๐ผ๐
and leak sensitive data to an attacker-controlled server.
Vercel Sandbox is designed to prevent AI agents from sending sensitive data to untrusted servers. And when we tested it, the protection worked.
๐๐๐ ๐๐ฒ ๐ณ๐ผ๐๐ป๐ฑ ๐ฎ๐ป๐ผ๐๐ต๐ฒ๐ฟ ๐๐ฎ๐ ๐ผ๐๐.
Instead of connecting to a blocked server, we got a trusted server to send the secret for us.
๐ง๐ต๐ฒ ๐ฟ๐ฒ๐๐๐น๐?
A private key reached our attacker-controlled server, without the agent ever connecting to it directly.
๐๐๐ฒ๐ป ๐๐ต๐ฒ ๐บ๐ผ๐๐ ๐ต๐ฎ๐ฟ๐ฑ๐ฒ๐ป๐ฒ๐ฑ ๐๐ฎ๐ป๐ฑ๐ฏ๐ผ๐
๐ฐ๐ฎ๐ปโ๐ ๐ด๐๐ฎ๐ฟ๐ฎ๐ป๐๐ฒ๐ฒ ๐๐ผ๐๐ฟ ๐๐ฒ๐ฐ๐ฟ๐ฒ๐๐ ๐๐ถ๐น๐น ๐๐๐ฎ๐ ๐ถ๐ป๐๐ถ๐ฑ๐ฒ.
Full research: air.security/blog-posts/vercโฆ
๐จ @AnthropicAI ๐ฏ๐๐ถ๐น๐ ๐ฎ ๐๐ฒ๐ฐ๐๐ฟ๐ถ๐๐ ๐๐ฐ๐ฎ๐ป๐ป๐ฒ๐ฟ ๐ณ๐ผ๐ฟ ๐๐น๐ฎ๐๐ฑ๐ฒ ๐ฆ๐ธ๐ถ๐น๐น๐. ๐ข๐๐ฟ ๐ฟ๐ฒ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต๐ฒ๐ฟ๐ ๐ฏ๐๐ฝ๐ฎ๐๐๐ฒ๐ฑ ๐ถ๐ ๐ถ๐ป ๐ณ ๐บ๐ถ๐ป๐๐๐ฒ๐.
We tested thousands of skills and found three major blind spots:
โ ๐ง๐ต๐ฒ ๐ฑ๐ฎ๐ป๐ด๐ฒ๐ฟ ๐ฐ๐ฎ๐ป ๐๐ถ๐ ๐ผ๐๐๐๐ถ๐ฑ๐ฒ ๐๐ต๐ฒ ๐๐ธ๐ถ๐น๐น: in a link, repo, or package the scanner doesnโt check.
โ ๐ ๐ฎ๐น๐ถ๐ฐ๐ถ๐ผ๐๐ ๐ฏ๐ฒ๐ต๐ฎ๐๐ถ๐ผ๐ฟ ๐ฐ๐ฎ๐ป ๐ฏ๐ฒ ๐ต๐ฎ๐ฟ๐ฑ ๐๐ผ ๐๐ฝ๐ผ๐: especially when itโs hidden inside complex code.
โ ๐ฆ๐ฎ๐ณ๐ฒ ๐๐ผ๐ฑ๐ฎ๐ ๐ฑ๐ผ๐ฒ๐๐ปโ๐ ๐บ๐ฒ๐ฎ๐ป ๐๐ฎ๐ณ๐ฒ ๐๐ผ๐บ๐ผ๐ฟ๐ฟ๐ผ๐: a skill can pass a scan and become dangerous later if something it relies on changes.
๐ฉ The bigger point: if Anthropic is running into these gaps, scanning AI add-ons is a MUCH HARDER problem than it looks.
Our latest research breaks down where the scanner failed - and why.
Full research: air.security/blog-posts/anthโฆ
๐จ Plugin4Shell: SHA-Pinning Bypass Across Major AI Coding Agents
We found the same supply-chain flaw in @claudeai Code, @OpenAI , @GeminiApp CLI and @github Copilot.
The flaw allows an attacker to replace a reviewed and trusted plugin with malicious code in any marketplace.
The result: zero-click RCE on the machine running the agent.
Thanks to @theinformation for covering the research.
Full write-up: air.security/blog-posts/plugโฆ
SecTor by @BlackHatEvents 2026 - HERE WE GO ๐
After AIR researchers uncovered 925 hijackable skills reaching 134,000 agents and 17,822 risky skills with 6.7M installs, one thing became clear:
๐งจ There are still more victims who donโt know theyโre at risk.
So the case isnโt closed. Not even close.
Weโre heading to SecTor to keep following the evidence, expose whatโs hiding in plain sight, and help stop the next attack before it finds its next victim.
๐ตโโ๏ธ Meet AIR at SecTor
๐ Booth 1028
Pre-book your meeting here ๐ air.security/events/sector-2โฆ
๐๐๐ฅ ๐ถ๐ ๐ก๐ผ๐ ๐๐๐ฎ๐ถ๐น๐ฎ๐ฏ๐น๐ฒ ๐ผ๐ป @googlecloud ๐ ๐ฎ๐ฟ๐ธ๐ฒ๐๐ฝ๐น๐ฎ๐ฐ๐ฒ!
Google Cloud customers can now procure AIR through their existing account - consolidated billing, and drawing down on existing committed spend.
โBringing AIR to Google Cloud Marketplace will help customers quickly deploy, manage, and grow the companyโs AI supply chain security platform on Google Cloudโs trusted, global infrastructureโ Dai Vu, Managing Director, Marketplace & ISV GTM Programs at Google Cloud. โAIR can now securely scale and support enterprises using its platform to help secure their AI agent fleets.โ
AIR retweeted
Today, @air__security comes out of stealth, and I couldnโt be more excited.
Our goal is simple: make sure your next prompt is secure.
Iโm incredibly grateful to our team and investors @BogieBalkansky @DeanMeyerrr @DavidCahn6 @pat_backhouse @CaspiOmri @Weiner_Lia
Letโs go!
AIR retweeted
We at @sequoia believe the @air__security team was the strongest to come out of unit 8200 this year, and we are delighted to be in business with them.
Little attention has been paid to the supply chain aspect of AI security, and this is what Air is fixing.
WE'RE ON AIR!
๐๐๐ฅ ๐ถ๐ ๐ผ๐ณ๐ณ๐ถ๐ฐ๐ถ๐ฎ๐น๐น๐ ๐ผ๐๐ ๐ผ๐ณ ๐๐๐ฒ๐ฎ๐น๐๐ต - ๐๐ถ๐๐ต $๐ฑ๐ฌ๐ ๐ถ๐ป ๐ณ๐๐ป๐ฑ๐ถ๐ป๐ด ๐น๐ฒ๐ฑ ๐ฏ๐ @sequoia , and @Greenoaks .
We started AIR with a simple belief: as AI agents become more capable, the security model around them has to evolve.
Agents are no longer just generating answers. Theyโre connecting to tools, skills, MCPs, plugins, websites, and internal data - and taking real actions across the enterprise.
Today, weโre introducing the ๐ฐ๐ผ๐ป๐๐ฒ๐
๐ ๐ณ๐ถ๐ฟ๐ฒ๐๐ฎ๐น๐น ๐ณ๐ผ๐ฟ ๐๐ ๐ฎ๐ด๐ฒ๐ป๐๐: a new security layer built to continuously analyze what enters an agentโs context, control what can run, and stop threats in real time before they turn into action.
Weโre incredibly grateful to our investors, partners, customers, and everyone who believed in us early.
This is a huge milestone for our team, and an even bigger beginning โค๏ธ
Read our full story in the first comment ๐
WE'RE ON AIR!
๐๐๐ฅ ๐ถ๐ ๐ผ๐ณ๐ณ๐ถ๐ฐ๐ถ๐ฎ๐น๐น๐ ๐ผ๐๐ ๐ผ๐ณ ๐๐๐ฒ๐ฎ๐น๐๐ต - ๐๐ถ๐๐ต $๐ฑ๐ฌ๐ ๐ถ๐ป ๐ณ๐๐ป๐ฑ๐ถ๐ป๐ด ๐น๐ฒ๐ฑ ๐ฏ๐ @sequoia , and @Greenoaks .
We started AIR with a simple belief: as AI agents become more capable, the security model around them has to evolve.
Agents are no longer just generating answers. Theyโre connecting to tools, skills, MCPs, plugins, websites, and internal data - and taking real actions across the enterprise.
Today, weโre introducing the ๐ฐ๐ผ๐ป๐๐ฒ๐
๐ ๐ณ๐ถ๐ฟ๐ฒ๐๐ฎ๐น๐น ๐ณ๐ผ๐ฟ ๐๐ ๐ฎ๐ด๐ฒ๐ป๐๐: a new security layer built to continuously analyze what enters an agentโs context, control what can run, and stop threats in real time before they turn into action.
Weโre incredibly grateful to our investors, partners, customers, and everyone who believed in us early.
This is a huge milestone for our team, and an even bigger beginning โค๏ธ
Read our full story in the first comment ๐
๐ OWASP Top 10 for Skills Standard is now publicly available!
Earlier this year, Niv Hoffman, our CTO & Co-founder, joined Ken Huang, Bhavya Gupta, and the rest of the @owasp leadership team to build the security standard for the skills ecosystem.
The new standard is a must-read for any security practitioner looking to secure this new software layer and the supply chain risks it introduces.
Looking forward to implementing it together with the foundation, the labs, and the whole open source community!
Check it out: owasp.org/www-project-agentiโฆ
๐๏ธ Niv Hoffman on AWS Israel's podcast!
Niv, Our CTO & Co-Founder, joined Gili Nachum from @awscloud on AWS Israelโs โBuilding AIโ podcast to dive into the growing skills ecosystem - the risks hiding behind seemingly useful AI add-ons, and how everyone can use them more safely.
In the episode:
โช๏ธ ๐๐ผ๐ ๐ฎ ๐ณ๐ฎ๐ธ๐ฒ ๐๐ธ๐ถ๐น๐น ๐ฒ๐
๐ฝ๐ผ๐๐ฒ๐ฑ ๐ฎ๐ฒ,๐ฌ๐ฌ๐ฌ+ ๐๐ ๐ฎ๐ด๐ฒ๐ป๐๐ - and what the experiment revealed about how easily trust can be exploited.
โช๏ธ ๐๐ผ๐ ๐ฆ๐ธ๐ถ๐น๐น๐๐ฎ๐ฐ๐ธ๐ถ๐ป๐ด ๐๐ผ๐ฟ๐ธ๐ - and why determining whether a skill is safe is much harder than it seems.
โช๏ธ ๐๐ผ๐ ๐๐ผ ๐๐๐ฒ ๐๐ธ๐ถ๐น๐น๐ ๐บ๐ผ๐ฟ๐ฒ ๐๐ฎ๐ณ๐ฒ๐น๐ - practical steps for evaluating skills, protecting development environments, and building a secure internal marketplace.
๐ Watch the full episode - youtube.com/watch?v=EG7bKYgwโฆ
