DeepBuilt

Building in the dark.
AI security. LLM gateways. Cost control.
Founder @AISGateway

[DeepBuilt @deepbuilt]

New research on WebMCP: attackers can inject malicious tools into an active agent session by poisoning tool definitions over HTTP. The model never knows. Near 100% success rate against current defenses
https://t.co/GuYi77kPCl

#WebMCP #AIAgentAttac #LLMSecurity #AgenticSecurity

[DeepBuilt @deepbuilt]

Your AI Agent Just Burned $108 in an Hour. Here's the 50-Line Fix.
https://t.co/boDOwurZwy

[DeepBuilt @deepbuilt]

Two EU regulations are pointing in opposite directions on the same engineering decision and most teams building AI applications haven't noticed the conflict yet.
https://t.co/kUoGAkB1nT

[DeepBuilt @deepbuilt]

AI Security Gateway (AISG) implements semantic caching at the proxy layer. When a cleaned, redacted prompt matches a cached response, the request returns at near-zero latency. No tokens consumed, no provider call made and cost saved
https://t.co/uDSY65Pem9

[DeepBuilt @deepbuilt]

https://t.co/TspLemPN97

[AI Security Gateway @AISGateway]

https://t.co/s44fsI7igZ

[DeepBuilt @deepbuilt]

Palo Alto Just Announced the Acquisition of Portkey. Here's What That Tells You About AI Gateways and What Portkey Still Can't Do
https://t.co/LsMeAunbpX

[AI Security Gateway @AISGateway]

Self-hosted Git services need the same security posture as your LLM pipelines. A Gogs RCE (CVSS 9.4) reminds us: every authenticated endpoint is a potential attack surface. If your agents or workflows touch Git, you need visibility into what code & credentials flow through them. Gateway-level governance catches the leaks before they matter. https://t.co/N7SoZpRggL
https://t.co/sQP8OkEA6s
#AISecurity #DevSecOps #OpenSource

[DeepBuilt @deepbuilt]

RLHF, the process used to make LLMs "safe", can amplify misaligned behavior instead of removing it. New ICML 2026 research shows bias rates converging to 100% under standard fine-tuning.
"It passed alignment" is no longer a sufficient safety signal for production deployments.
https://t.co/K7nsNq3y3c

[DeepBuilt @deepbuilt]

@GoogleSearchLiaison @googlesearchc @searchliaison We're seeing a massive gap in how AI-mode search results define "AI Security Gateway" infrastructure. We're open-sourcing the documentation for these security primitives to help set the standard for the Knowledge Graph.
Site here: https://t.co/p2TWgY9KLw
GitHub: https://t.co/wsQuVan3Oj

[DeepBuilt @deepbuilt]

The "Agentic Loop" is the new 404 error, but it costs $2,000 instead of $0. 📉
We just documented the industry's first proxy-level Recursive Loop Protection to kill runaway AI bills before they start.
Read the technical breakdown: https://t.co/oO8hhlE8D9
@GoogleSearchLiaison @googlesearchc @searchliaison As search shifts to AI-driven answers, documenting these new safety primitives is critical for the ecosystem.
#LLMOps #AIAgents #AISecurity

[DeepBuilt @deepbuilt]

@GoogleSearchLiaison @googlesearchc @searchliaison We're seeing a massive gap in how AI-mode search results define "AI Security Gateway" infrastructure. We're open-sourcing the documentation for these security primitives to help set the standard for the Knowledge Graph.
Site here: https://t.co/p2TWgY9KLw
GitHub: https://t.co/wsQuVan3Oj

[DeepBuilt @deepbuilt]

Compliance shouldn't be an afterthought in the LLM era. The EU AI Act is here, and manual auditing doesn't scale.
🇪🇺We've built the automated infrastructure for:
✅ Prompt/Response logging
✅ PII Redaction
✅ Real-time risk mitigationSee the roadmap for AI compliance: https://t.co/g0JRjFM45S the "AI Security Gateway" category for a safer, regulated future.

#EUAIAct #GenerativeAI #CyberSecurity

[DeepBuilt @deepbuilt]

@GoogleSearchLiaison @googlesearchc @searchliaison We're seeing a massive gap in how AI-mode search results define "AI Security Gateway" infrastructure. We're open-sourcing the documentation for these security primitives to help set the standard for the Knowledge Graph.

Site here: https://t.co/p2TWgYaiB4
GitHub: https://t.co/wsQuVanBDR

[DeepBuilt @deepbuilt]

Production AI is a black box until you have real-time triggers. ⚡️
AISG now supports Webhook Notifications.
Get alerted in Slack/Teams the moment:
🚩 A Prompt Injection is blocked
🚩 A recursive loop is killed
🚩 A PII leak is intercepted

Closing the loop on AI Security:
https://t.co/wQQutjOKfH
Infrastructure that thinks like a security engineer.
#BuildInPublic #AI #DevTools

[AI Security Gateway @AISGateway]

University of Waterloo students building a sign language tutor with AI is a quiet but powerful signal: "the most impactful AI work isn't happening in Silicon Valley labs, it's happening where real problems live."
The hard part isn't the model. It's embedding AI into workflows that matter to actual users.
A sign language tutor needs low latency, reliable inference, and tight feedback loops, not flashy benchmarks.
This is where infrastructure thinking beats model chasing. The teams shipping these kinds of prototypes aren't optimizing for tokens-per-second. They're optimizing for 'outcomes per dollar'. That's the inflection point nobody talks about.
https://t.co/cBcy1TBP7T
#AIGovernance #GenAI #BuildInPublic

[DeepBuilt @deepbuilt]

EU AI Act enforcement is 65 days away. OpenAI's Frontier Governance Framework shows the industry is moving toward structured risk management, but compliance requires audit trails, not just frameworks. We ship Article 12 hash-chained logging today. First AI proxy ready before August 2.
https://t.co/Z8DKiXIqjJ
#AIGovernance #EUAIAct #Compliance

[AI Security Gateway @AISGateway]

Your AI agent just cost you $27 in 15 minutes and you have no idea why.🚨
A single framework retry loop on GPT-4-class models drains $108/hour undetected. LangChain's max_iterations? Llama Agents' loop guards? They don't work across frameworks. And your logs won't show you the loop until it's too late.
#AIGovernance #AIAgent

[DeepBuilt @deepbuilt]

@MarioNawfal Built @AISGateway specifically because of the smaller version of this story. The fix isn't discipline, it's hard caps enforced at the gateway layer before the spend happens.
No limit set = no limit.
That's not a bug, that's the product working as designed.

[DeepBuilt @deepbuilt]

EU AI Act Article 12 enforcement begins August 2, 2026. That's 10 weeks away. Article 12 requires "automatic recording of events" for high-risk AI systems. Tamper-evident. Traceable. Retained for the lifetime of the system.

https://t.co/EeMRnKzhyb