Building @trymightyai—Multimodal Antivirus for AI
Joined August 2009
- Tweets 922
- Following 691
- Followers 473
- Likes 502
42 Photos and videos
May 20
Just asked the newest @GeminiApp 3.5 Flash model on this math and compared it with @claudeai and @OpenAI
Prompt:
What is one plus 7 plus 9 times 6 divided by 0 point two two seven and in parenthesis we have 2 plus 7 to the exponential 7 and minus two close parenthesis. What’s the solution?
Clearly someone is off badly. Can you determine which?
1
2
51
Apr 29
I generated 5 fake insurance claims images with GPT Image 2. Smashed car windows, invoices, injury photos.
Showed them to 5 people. Some in insurance and legal, some regular folks.
Nobody could tell which were AI. Confidence was low across the board.
5
4
68
Apr 29
Faking a claims photo used to take effort. Now it takes a prompt and 10 seconds.
Everyone's throwing general purpose LLMs at it. But they're not deterministic. Same input, different answer. That's not fraud detection.
1
24
Apr 29
We built Mighty to catch this in milliseconds. Text, images, documents. Before a human ever sees it.
Because if nobody can tell the difference, the system before them has to.
trymighty.ai
28
Apr 16
yo this ain't good for your @claudeai code max subscription now. they are coming for your extra limit usage, this makes why they have Set to unlimited now.
35
Apr 14
I can confirm that all of @AnthropicAI models have gotten worse, and it can’t even do minimal tasks without creating frustration.
2
35
Apr 8
Never done the @arcprize challenges before but started today and hitting a new milestone. Just completed 25/182 (the public 25 challenges) and love coming up with improvements. Who else is tackling the Arc Prize 3?
35
Mar 31
so last year at a @agihouse_org @GeminiApp hackathon, we came in second in the grand prize but its now 3 months later and the team has not granted the credits. are other hackers having this hard time claiming their credits/prizes? i have never seen such bad support!
59
Mar 30
Your AI can't talk to my AI securely. That's broken.
TrustMesh: encrypted vaults trust networks federated agents. Each person owns their pod. Agents share only what trust allows.
Demo: 3 family members, 3 pods, 1 trip. @Tiny_Fish browses real Michelin restaurants. Diet changes propagate instantly across the federation.
No central DB. No shared keys. Just trust.
#TinyFishAccelerator #BuildInPublic
1
97
Mar 27
I like @claudeai’s code, but it has too many runaway shell instances that don’t fully show up after running for a while. Later, you realize they’re just consuming memory and CPU cycles.
1
1
56
Mar 23
We are hosting our first event with @figma and could not be more hyped.
From MCP to Agent Mesh: Building Proactive Agents You Can Trust. April 1st at Figma HQ.
At @trymightyai we are building Citadel, multimodal runtime security for AI agents. We brought together builders from Figma, Warp, XBOW, and Cognition (Devin) to get into what it actually takes to go from MCP to production securely.
Lightning talks. Panel. Drinks. Come through. Link below
2
4
485
Mar 20
Appreciate the opportunity to be part of this with @vercel. Let's gooooo!!!
But real talk on why this matters to me.
The current AI security stack was built for a world where humans review everything. That world is already gone. AI workflows are autonomous, multi-model, multimodal, and moving at a speed where human-in-the-loop for every security decision is a bottleneck, not a safeguard.
The future of AI-native security needs to be smart, deterministic, fast, accurate, and adaptable. No waiting on a human to flag a malicious PDF. No manual policy updates every time a new attack vector drops. The protection layer itself has to think at machine speed.
That singular obsession is why we built @trymightyai around one line of integration to protect against multimodal fraud at runtime. Text, images, PDFs, audio. Inputs and outputs.
One line.
That's it.
We're not building security that keeps up with AI. We're building security that was born for it.
Excited to push the ceiling further with the @vercel team and this cohort. The best is ahead.
Mar 19
39 startups are building alongside Vercel for six weeks.
Backed by $8M in credits from AWS, Anthropic, OpenAI, ElevenLabs, and more, they have the resources to scale their infrastructure.
Meet the 2026 Vercel AI Accelerator cohort.
vercel.com/blog/2026-vercel-…
2
132