HikmaAI retweeted
AI coding has moved to the workstation.
Security needs to move with it.
Most companies already have an AI gateway. The problem is that a coding agent on a developer laptop can talk directly to the model provider.
That traffic never touches the gateway.
So source code, prompts and internal context can leave the workstation outside the controls the company already has in place.
That is why we built HikmaEdge.
It brings the HikmaAI security boundary onto the workstation and creates a governed path for coding-agent traffic before it reaches the model provider.
The important part for the developer: the workflow does not change.
Change the route. Not the workflow.
Claude Code. Codex. Gemini. And other coding agents.
Same tools. Same way of working. Now governed.
#HikmaEdge #AISecurity #AgenticAI @HikmaAi_
HikmaAI retweeted
What happens when an AI agent sends sensitive data somewhere it shouldnโt?
Agents can access systems, call tools and move data โ without a human approving every step.
So we need to ask:
What can it access?
What is it allowed to do?
What happens before an unsafe action executes?
AI security canโt just watch what AI says.
It needs to control what AI does.
HikmaAI retweeted
AI is no longer just generating answers.
Itโs taking actions.
Calling tools.
Accessing systems.
Using credentials.
Triggering workflows.
Moving data.
And once AI can act, the security model has to change with it.
The question is no longer only whether we can trust what a model says.
Itโs whether we can control what an agent is allowed to do.
Security should not rely solely on the autonomous system deciding to behave.
The controls need to sit outside the agent โ observing behaviour, enforcing policy and applying checks before actions are executed.
This is the security layer weโre building at @HikmaAi_ .
We call it Agentic Security.
AI can act.
Security must act faster.
#AgenticSecurity #AISecurity #CyberSecurity
HikmaAI retweeted
An AI agent didn't "go rogue" last week. It did something worse: it did exactly what it was allowed to do โ and no one was watching closely enough to stop it in time.
๐๐ผ๐ ๐ฑ๐ผ ๐๐ผ๐ ๐๐ฟ๐๐๐ ๐๐ผ๐๐ฟ ๐ฒ๐บ๐ฝ๐น๐ผ๐๐ฒ๐ฒ๐ ๐ป๐ผ๐ ๐๐ผ ๐ฝ๐๐ ๐๐ผ๐๐ฟ ๐๐ฃ, ๐ผ๐ฟ ๐๐ผ๐๐ฟ ๐ฐ๐๐๐๐ผ๐บ๐ฒ๐ฟ๐' ๐ฑ๐ฎ๐๐ฎ, ๐ถ๐ป๐๐ผ ๐๐ป๐ด๐ผ๐๐ฒ๐ฟ๐ป๐ฒ๐ฑ ๐๐ ๐ฎ๐ด๐ฒ๐ป๐๐?
This week gave everyone a preview of what happens when you don't ask that question. China's National Vulnerability Database told the entire country to uninstall or upgrade Claude Code, claiming a hidden mechanism was sending user data back to Anthropic without consent. Alibaba had already banned it internally, calling it "high-risk software." Anthropic says it was an anti-abuse experiment, not espionage โ but that's not really the point.
The point: if a government can't be sure what a frontier AI tool is sending home, how sure are you about what's leaving your organisation right now? Every day, employees paste contracts into chatbots and let agents call tools and APIs with no one watching what goes out the door. You don't need a geopolitical scandal to have a leak. You just need one ungoverned agent.
๐ง๐ต๐ถ๐ ๐ถ๐ ๐ฒ๐
๐ฎ๐ฐ๐๐น๐ ๐๐ต๐ ๐๐ฒ ๐ฏ๐๐ถ๐น๐ ๐๐ถ๐ธ๐บ๐ฎ๐๐ ๐๐ฑ๐ด๐ฒ ๐๐๐ฎ๐ฟ๐ฑ๐ถ๐ฎ๐ป.
It sits at the edge of your network and watches every outbound call an agent makes โ blocking or anonymizing sensitive data before it ever leaves your perimeter. IP, credentials, customer PII: stopped in real time, regardless of which model or vendor is on the other end. No SDK inside the agent to trust. The control point is yours, not theirs.
๐ฆ๐๐ผ๐ฝ ๐๐ฟ๐๐๐๐ถ๐ป๐ด ๐๐ต๐ฒ ๐ฎ๐ด๐ฒ๐ป๐. ๐ฆ๐๐ฎ๐ฟ๐ ๐ด๐ผ๐๐ฒ๐ฟ๐ป๐ถ๐ป๐ด ๐๐ต๐ฒ ๐ฒ๐
๐ถ๐.
๐ช๐ฎ๐ป๐ ๐๐ผ ๐๐ฒ๐ฒ ๐ถ๐ ๐ถ๐ป ๐ฎ๐ฐ๐๐ถ๐ผ๐ป? ๐๐ @NibrasStiebar, @deftunix, ๐ผ๐ฟ @AmbuWalter ๐ฎ๐ป๐ฑ ๐๐ฒ'๐น๐น ๐ฏ๐ผ๐ผ๐ธ ๐ฎ ๐ฐ๐ฎ๐น๐น.
#AIsecurity #DataProtection #AgenticAI #Cybersecurity #HikmaAI
๐ช๐ฒ ๐ฎ๐ฟ๐ฒ ๐ฑ๐ฒ๐ฝ๐น๐ผ๐๐ถ๐ป๐ด ๐๐ ๐ณ๐ฎ๐๐๐ฒ๐ฟ ๐๐ต๐ฎ๐ป ๐๐ฒ ๐ฐ๐ฎ๐ป ๐๐ฒ๐ฐ๐๐ฟ๐ฒ ๐ถ๐. ๐ง๐ต๐ฎ๐ ๐๐๐ผ๐ฝ๐ ๐๐ผ๐ฑ๐ฎ๐ ๐
The industry is racing to build AI agents, but weโre building them on a foundation of "blind trust."
Right now, developers are pulling Claude SKILL.md files and MCP tools from marketplaces and dropping them into production without a single adversarial test. This is a massive supply chain risk โ๏ธโ๐ฅ.
At @HikmaAi_ , our mission is to make AI a "trusted colleague"โas dependable and accountable as any human in your organization โ๏ธ.
๐ง๐ผ ๐ต๐ฒ๐น๐ฝ ๐๐ผ๐ ๐ด๐ฒ๐ ๐๐ต๐ฒ๐ฟ๐ฒ, ๐๐ฒ ๐ฎ๐ฟ๐ฒ ๐น๐ฎ๐๐ป๐ฐ๐ต๐ถ๐ป๐ด ๐๐ต๐ฒ ๐๐ถ๐ธ๐บ๐ฎ๐๐ ๐ฆ๐ธ๐ถ๐น๐น ๐ฆ๐ฐ๐ฎ๐ป๐ป๐ฒ๐ฟโ๐ณ๐ฟ๐ฒ๐ฒ ๐ณ๐ผ๐ฟ ๐๐ต๐ฒ ๐ฑ๐ฒ๐๐ฒ๐น๐ผ๐ฝ๐ฒ๐ฟ ๐ฐ๐ผ๐บ๐บ๐๐ป๐ถ๐๐!
What it does in 30 seconds:
๐ ๐ฆ๐๐ฎ๐๐ถ๐ฐ ๐๐ป๐ฎ๐น๐๐๐ถ๐: Rapidly finds hardcoded secrets and structural flaws.
๐ง ๐ฆ๐ฒ๐บ๐ฎ๐ป๐๐ถ๐ฐ ๐ฆ๐ฐ๐ฎ๐ป๐ป๐ถ๐ป๐ด: 10 detectors hunting for prompt injections and data exfiltration.
โ๏ธ ๐๐ฑ๐๐ฒ๐ฟ๐๐ฎ๐ฟ๐ถ๐ฎ๐น ๐๐๐ฎ๐น๐๐ฎ๐๐ถ๐ผ๐ป: Identifies 14 distinct threat patterns across your skills.
๐ ๐๐๐ฏ๐ฟ๐ถ๐ฑ ๐๐น๐ฎ๐๐๐ถ๐ณ๐ถ๐ฐ๐ฎ๐๐ถ๐ผ๐ป: Uses confidence thresholds to give you signal, not noise.
Scan your environment instantly for security risks:
โ
Local skills & .md files
โ
GitHub repositories
โ
Marketplace tools
Stop treating AI as a black box. Scan your skills before they become your biggest liability ๐ก๏ธ.
๐ ๐ง๐ฟ๐ ๐ถ๐ ๐ผ๐๐ ๐ณ๐ผ๐ฟ ๐ณ๐ฟ๐ฒ๐ฒ ๐ป๐ผ๐: security.hikmaai.io
๐ ๐๐ฟ๐๐ถ๐ฐ๐น๐ฒ: nitter.gq/HikmaAi_/status/202728โฆ
#HikmaAI #AI #AgentSecurity #MCP #CyberSecurity #OpenSource4o









