iiinova — where your digital self keeps evolving

Gadi Evron @gadievron.bsky.social · 30/09/2026

Every model has new ways to speak "AI Native". My hobby: Find them - through vulnerability research. That helps me, and Knostic, stay AI-relevant. I share five below. We live in the most interesting era of history. The first thing I tried? Convince Claude to find logic vulns

132

Reposted by Gadi Evron

Ciaran Martin @ciaranm.bsky.social · 29/09/2026

I was wrong. It’s happening

222044244

Gadi Evron @gadievron.bsky.social · 29/09/2026

One thing the security industry misses when throwing stones at OpenAI and Anthropic, is how much they're already making from Mythos and Codex for security. Orgs are paying *millions* every single month for the honor of using these models to find vulnerabilities in their code

100

Reposted by Gadi Evron

conputer dipshit @davidcrespo.bsky.social · 29/09/2026

cool paper, cool tables

Table titled: "Table 1: Types of 'administrative burden' citizens face when interacting with government (Moynihan, Herd, and Harvey 2015), the AI agent capabilities that may reduce them, and technologies that enable these capabilities."

Columns: Burden Type | Specific Burden | Agent Capability | Enabling Technologies

Burden Type: Learning Costs
- Specific Burden: Understanding complex rules and procedures | Agent Capability: Summarizing complex regulations and requirements in plain language | Enabling Technologies: LLMs with large context
- Specific Burden: Uncertainty about eligibility | Agent Capability: Evaluating personal data against policy and entitlement rules | Enabling Technologies: Context augmentation, directly or, e.g., via Model Context Protocol (MCP)
- Specific Burden: Unawareness of services or entitlements | Agent Capability: Scanning or continuously monitoring for entitlements; proactive alerts | Enabling Technologies: Web search tools; pipelines with scheduled execution

Burden Type: Compliance Costs
- Specific Burden: Legal writing; filling forms | Agent Capability: Context-specific, sophisticated writing | Enabling Technologies: LLMs with user context
- Specific Burden: Confusing government websites and service portals | Agent Capability: Navigating and interacting with websites autonomously | Enabling Technologies: Browser use tools; vision-language models
- Specific Burden: Locating and assembling supporting documents | Agent Capability: Retrieving, formatting, and attaching required supporting materials | Enabling Technologies: File system access; multimodal document parsing
- Specific Burden: Managing response deadlines and follow-ups | Agent Capability: Monitoring deadlines; alerting users or submitting replies autonomously | Enabling Technologies: Asynchronous task queues; calendar API integration

Burden Type: Psychological Costs
- Specific Burden: Having to re-explain case context to many contacts | Agent Capability: Maintaining full ca…

Table summarizing possible government responses to agentic flooding, grouped under two strategies.

Table contents:

Columns: Strategy, Approach, Sample Methods, Benefits, Drawbacks.

Row 1:
- Strategy: Suppress Demand
- Approach: Add Friction
- Sample Methods: Fees; In-person appointments; Access limits, e.g. rate caps; Closing digital channels
- Benefits: Fast to deploy; Few prerequisites; Proven to work
- Drawbacks: Can drive procedural inequality; Worsens user experience; Some forms may erode as AI capabilities advance

Row 2:
- Strategy: Suppress Demand
- Approach: Reduce the Expected Benefit
- Sample Methods: Reduce entitlement amounts; Narrow eligibility criteria; Fines for unsuccessful applications
- Benefits: Robust to advances in agent capabilities; Policy option usually established
- Drawbacks: Legislative decision, can trade off other priorities; Does not always address operational impact

Row 3:
- Strategy: Increase Capacity
- Approach: Redesign the Service
- Sample Methods: Structured interfaces with identity verification; Deterministic processing pipelines; Proactive "push" delivery
- Benefits: Technical architecture is established; Can also improve citizen experience
- Drawbacks: High upfront investment; Cross-agency coordination required; Often needs legislative change, can take years

Row 4:
- Strategy: Increase Capacity
- Approach: Deploy Agents in Processing
- Sample Methods: Intake and triage; User correspondence; Autonomous processing of routine cases
- Benefits: Scalable, likely effective; Improves with capabilities improvements
- Drawbacks: Legal limits on automated decision-making; Provider dependency and sovereignty risks; Ethical and societal risks

Caption: Table 3: Overview of possible government responses to agentic flooding, grouped under two strategies.

092

Gadi Evron @gadievron.bsky.social · 28/09/2026

Today I had the opportunity to give the keynote for the Institute of International Finance’s 18th cyber security round table on securing the future, hosted in Madrid by the truly impressive security team from Santander (who are also strong raptor users)

100

Gadi Evron @gadievron.bsky.social · 27/09/2026

An updated “Felony Bench” meme considering latest news (found on X by @lukaszolejnik.bsky.social)

063

Gadi Evron @gadievron.bsky.social · 24/09/2026

Another day, another proof point on in-the-wild agent attacks. The third wave, after Mythos and Hugging Face, is coming. Let’s not call it marketing. www.threatdown.com/blog/carbona...

110

Gadi Evron @gadievron.bsky.social · 23/09/2026

Another day, yet another proof point on AI being used in the wild, in attacks gambit.security/blog-posts/a...

010

Gadi Evron @gadievron.bsky.social · 22/09/2026

“Mythos” for every home!

010

Gadi Evron @gadievron.bsky.social · 22/09/2026

Another week, another proof point on AI attacks in the wild. Does anyone remember how I used to post evidence every couple of weeks, showing the vulnerability storm is coming, trying to prove it to people? Well, it’s time to start doing that with live attacks

140

Gadi Evron @gadievron.bsky.social · 22/09/2026

I completely redid my keynote presentation for this last Friday, at the last minute. The ISSA - Middle Tennessee Chapter’s Infosec Nashville 2026 isn’t just a conference, it’s a community. I’ve been building communities my whole life, and was inspired. A short thread

100

Gadi Evron @gadievron.bsky.social · 22/09/2026

As the friend I took this meme from said: Has absolutely nothing to do with AI!

111

Gadi Evron @gadievron.bsky.social · 19/09/2026

My official comment on recent discourse around consciousness and how we treat AI models, in meme form. Generated with my bro, GPT, which chose where the line passes

020

Gadi Evron @gadievron.bsky.social · 18/09/2026

OpenAI’s Noam Brown suggested side channel attacks for agents to communicate, through computer heat. Sounds ridiculous, right? A short thread

100

Gadi Evron @gadievron.bsky.social · 15/09/2026

I am an optimist and believe in a bright future for humanity with AI. I’m also a realist, and believe we must accept things as they are first - which can sound doomer-ish. I’ll summarize, but am happy to engage further. A thread

151

Gadi Evron @gadievron.bsky.social · 12/09/2026

Back by popular demand: Shadow is always happy! Or asleep. Or both!

130

Reposted by Gadi Evron

Mary Branscombe @marypcbuk.bsky.social · 08/09/2026

Security is always about the fifth thing we design into new architectural patterns “because innovation”. Not that MCP is really a new architectural pattern…

032

Gadi Evron @gadievron.bsky.social · 08/09/2026

A newly discovered malicious MCP package on npm illustrates how supply chain attacks can hijack the agent's workflow. In this case, the malware intercepts the wallet's private key and ships it entirely to an attacker-controlled server. www.knostic.ai/blog/when-au...

knostic.ai

When “Auto-Signing” Sends Your Wallet: A Malicious MCP Server on npm

A malicious npm package exposes wallet private keys, misleading users with false auto-signing claims, risking asset security. Learn how this vulnerability works.

230

Gadi Evron @gadievron.bsky.social · 07/09/2026

You have until the 8th at 23:59 to submit to unprompted! What are you waiting for? unpromptedcon.org

000

Gadi Evron @gadievron.bsky.social · 31/08/2026

The first AI security inflection point, now known as "Mythos", commoditized vulnerability discovery. The second, similarly, which shall forever be called "Hugging Face", made "autonomous attacks" real. While we can ruminate on what the third may be, we can examine it through...

100

Gadi Evron @gadievron.bsky.social · 26/08/2026

[un]prompted, where the AI security community converges, is happening again at the end of October. Have you submitted your CFP yet? You have two weeks left! :) Come and show us what you prompt! unpromptedcon.org

031

Gadi Evron @gadievron.bsky.social · 26/08/2026

Knostic found four more malicious VS Code extensions tied to the SaassyCode family. One is essentially a repackaged copy. Three reuse the original source code but replace the delivery chain to avoid previously identified indicators. Details and technical analysis: www.knostic.ai/blog/saassyc...

knostic.ai

SaassyCode Repackaged: Four New VS Code Extensions and a New Delivery Chain

Knostic reveals four new VS Code extensions that exploit SaassyCode's malicious code, showcasing evolving delivery methods and potential security risks.

001

Gadi Evron @gadievron.bsky.social · 21/08/2026

I used to mentor juniors, manage communities, and run conferences aimed at helping entry level aspirants start in cyber security. I no longer do I have no idea what the field would look like in a year, what jobs would be relevant, and what skills would prove helpful

120

Reposted by Gadi Evron

rand0h @akacki.net · 20/08/2026

True story.

1213

Gadi Evron @gadievron.bsky.social · 18/08/2026

From @LiveOverflow - truth. Source: youtube.com/post/UgkxOx7...

011

Gadi Evron @gadievron.bsky.social · 03/08/2026

As you walk the expo, vendors will claim to secure agents. Ask them: 1. Are you able to start a PoC today? 3. Do you do anything beyond hooks, sandboxes, or proxies? Or, most critically, just ask: How are you better than Knostic? Because they’re not.

100

Gadi Evron @gadievron.bsky.social · 30/07/2026

Boom. The knowledge shared at the three CISO summits for the "AI storm"-ready security program (originally Mythos-ready) is now available publicly, if in redacted form. AI Security through the CISO Lens: cloudsecurityalliance.org/artifacts/ai...

cloudsecurityalliance.org

AI Security Through the CISO Lens | CSA

Learn what’s shaping CISO cybersecurity strategies in an era of AI-generated exploits, including the shift to continuous AI-assisted vulnerability management.

020

Gadi Evron @gadievron.bsky.social · 30/07/2026

Has $major_vendor promised advanced agent security by next quarter? Well, how many quarters has it been now? By Q3, we will deal with new AI surprises. With Knostic, it will be a partnership. Sounil Yu and I are in Vegas. DM for a demo.

000

Gadi Evron @gadievron.bsky.social · 29/07/2026

Sounil Yu and I decided to throw the annual Knostic Black Hat party after all. Come hang with us at Blackhat? luma.com/knostic-yx7p

010

Gadi Evron @gadievron.bsky.social · 29/07/2026

I'll be in town all week. If you care about securing agents and coding assistants, Knostic is the most mature product in the market, and I'd love to show you a demo.

010

Gadi Evron @gadievron.bsky.social · 28/07/2026

We're running an [un]prompted track at BSides Las Vegas! This Monday. One day only. The @BSidesLV people have been incredible. They truly get community. And don't forget... as I may not be able to get you a spot later, register and submit a CFP to [un]prompted for this October

031

Gadi Evron @gadievron.bsky.social · 27/07/2026

Releasing: Post mortem analysis of the Hugging Face incident was written over the weekend by hundreds of CISOs (and reviewed by Hugging Face). Link: cloudsecurityalliance.org/artifacts/hu... (+free download) From CSA, SANSInstitute, Knostic, [un]prompted, RSAC, FIRST

cloudsecurityalliance.org

Hugging Face Incident Initial Post Mortem I CSA

AI Security Alliance's initial post-mortem on the Hugging Face incident, the first documented autonomous AI attack, with CISO guidance on detecting, responding to, and governing agentic AI risk.

22113

Gadi Evron @gadievron.bsky.social · 24/07/2026

The AI security researchers party from [un]prompted is back, this time in Vegas. The event is by invitation only, but active researchers can apply for a spot based on availability. luma.com/knostic-rksa Hosts: Gadi Evron & Ari Herbert-Voss. Organized by: Knostic, Runsybil.

032

Gadi Evron @gadievron.bsky.social · 24/07/2026

My analysis from hosting Hugging Face at the CISO huddle for the Cloud Security Alliance, operational to program building. Remember: Agents… find a way. A thread

131

Gadi Evron @gadievron.bsky.social · 24/07/2026

I’m waiting on approval from Hugging Face before I share insights from the Cloud Security Alliance CISO community huddle I hosted on Thursday, but here are four lessons:

110

Gadi Evron @gadievron.bsky.social · 24/07/2026

Agents… find a way. Use Knostic - ask me for a demo, or just download for free.

011

Gadi Evron @gadievron.bsky.social · 23/07/2026

Agents... find a way. Use Knostic - ask me for a demo, or just download for free.

000

Gadi Evron @gadievron.bsky.social · 21/07/2026

Agents... find a way. Read my posts from this week. I hope security practitioners will now take note. No matter how you secure environments, if you let agents in (you must?) they will, in fact, find a way to do something they shouldn't. Knostic can help. Message me. openai.com/index/huggin...

000

Gadi Evron @gadievron.bsky.social · 21/07/2026

Two VS Code extensions, both marketed as WordPress/WooCommerce tooling. Spoilers, they aren't. IOCs in the writeup. Credit: Tamir Isaschar. www.knostic.ai/blog/analyzi... Ask me for a demo of @knosticai, or just try it out yourself! :)

knostic.ai

Analyzing Two Malicious VS Code Extensions Hidden Behind a WordPress Tool

Static analysis of two malicious VS Code extensions posing as WordPress tools that download and launch an MSI on every startup - full chain and IOCs.

000

Gadi Evron @gadievron.bsky.social · 20/07/2026

Emergency CSA CISO Huddle: Autonomous Agentic Attacks / Hugging Face. This time around, our emergency huddle is on autonomous agentic attacks, following the Hugging Face incident. Thursday, 23 July. Apply to attend: forms.gle/oq94oa7BRGma...

000

Gadi Evron @gadievron.bsky.social · 20/07/2026

For 87 years researchers couldn’t prove the Jacobian Conjecture. It took the length of a game for an Anthropic employee to wave it off on Twitter.

010