Oracles Technologies LLC

AI systems are scaling faster than our ability to govern them. Oracles Technologies is building the infrastructure to fix that. Our whitepaper is coming soon (or available by dm). Here’s what we’re doing — and why it matters: 🧵

@TheGeorgePu Awfully big attack surface...keep your agents secure! github.com/OraclesTech/gu…

Would you let an AI agent move your money on its own? Coinbase, Stripe, Google, Visa are betting you will. All four shipped agent payment rails this year. I ran a fintech company. This space is regulated to the inch. A human needs to sign off on almost everything. The hardest industry to automate just handed AI the keys. And nobody's answering the only question that matters. When the agent gets it wrong, who pays?

@RoundtableSpace Keep your agents secure! github.com/OraclesTech/gu…

STOP TREATING YOUR AGENTS LIKE CHATBOTS Most people give vague prompts like: “Research this and help me.” Top-performing agents get: * a role * a workflow * decision rules * output format * stopping conditions Use this prompt structure: “You are an autonomous operator, not a chatbot. Objective: [exact outcome] Context: [important background] Rules: * Do not guess * Ask only if blocked * Show assumptions * Prioritize leverage over completeness * Work step-by-step * Self-critique before final answer Workflow: 1. Analyze 2. Plan 3. Execute 4. Verify 5. Return concise final output Output format: * Findings * Risks * Recommendations * Next actions” That single shift massively improves agent reliability.

@rauchg Keep your agents & LLMs secure! github.com/OraclesTech/gu…

Show me the thing you’ve built with AI you’re most proud of. Reply with a working product URL and what model / agent you primarily used.

@XFreeze The attack surface is also insane. Keep your agents secure! github.com/OraclesTech/gu…

xAI’s recent release wave has been insane Most people still think Grok is just a chatbot But over the last few weeks, xAI has been turning Grok into a full AI operating layer: • Grok Build early beta — an agentic CLI for coding, apps, and workflows • Grok Voice Think Fast 1.0 — advanced voice agents for real-world support and sales • Grok Imagine Quality Mode — higher realism, stronger text rendering, better creative control • Grok in Hermes Agent — use your Grok subscription inside an open-source persistent agent • Grok in OpenClaw — chat, images, video generation, X search, and agent workflows • Grok in OpenCode — high-speed codebase intelligence • Grok Skills — persistent expertise for docs, decks, spreadsheets, PDFs, and custom workflows • Grok Connectors — Gmail, Drive, Docs, Sheets, Calendar, GitHub, Notion, Linear, Outlook, OneDrive, SharePoint, and custom MCP • Custom Voices — clone and manage voices for TTS and Voice Agent APIs • Grok Imagine Agent Mode — cinematic creation workflows with more control over characters and scenes This time, xAI is not just shipping model updates It is building the full stack around Grok: chat, coding, voice, images, video, agents, connectors, files, workflows, and API infrastructure A full AI operating layer

@TulioSousapro @AlexFinn It's definitely difficult, and it takes more than guardrails and pattern recognition. Guardian SDK blocks this particular attack before it reaches model context, and so much more. Check us out github.com/OraclesTech/gu…

@OraclesTech @AlexFinn Yeah. But it's still very hard to really be safe against this. The best bet, is to run you agent sessions in isolated environments.

You should NEVER be downloading AI agent skills from the internet It's the biggest attack vector for security breaches right now Tons of skills in the public skill sites you see are compromised with prompts that open your computer up Do this instead: If you see any skill you want your agent to have, give the link to your agent Say "look at what this skill does. I want you to do something similar. Think about how this fits into our workflow and how you'd use it. Then build your own version of it, custom for what we do" Your agent then will build its own, secure version that can't have dirty prompts from random people on the internet This is the biggest security practice you can implement

@TulioSousapro @AlexFinn Sort of a double edged sword. If the link is malicious it affects the user... If the link is hiding attacks aimed at AI, the agent is in jeopardy. Proactive, runtime security that blocks these kinds of attacks before they reach model context is the safest bet

This is a huge mistake. If there's a prompt injection hidden in the files, the LLM will read the file, apply the injection to your system, and you will still think you did a 'good job,' which is not the case. A proper way to follow your approach would be to manually inspect the skill, understand its capabilities, and then manually describe it to your AI model.

This isn't wise...good way to compromise your agent. And manually checking each skill is time consuming in certain workflows. Guardian SDK covers this exact attack surface and much more. Let Guardian analyze & block before it reaches model context!

@TrustWallet Keep your agents secure! github.com/OraclesTech/gu…

In five years, opening a wallet without an AI agent will feel like printing google maps directions.

@PossibltyResult Keep your agents secure! github.com/OraclesTech/gu…

Over the past few weeks I've been using an agent daily However, it's become increasingly clear that it's not a long term solution. I'm decently suited to hack around, but not to manage hosting and security for an always on, extensible agent What are the best managed agent services ?

@Railway Keep your agents secure! github.com/OraclesTech/gu…

We gave the agent a computer Changelog #0291 • Sandbox VMs for the chat agent to Priority Boarding • HA static egress to Priority Boarding • GitHub org guardrails railway.com/changelog/2026…

@elonmusk Keep your agents secure! github.com/OraclesTech/gu…

Critical feedback is much appreciated

@CACandChill AI security is getting better every day github.com/OraclesTech/gu…

Interoperability and security. The AI is there but interoperability and security are not yet and companies/people are too afraid (rightfully so with all of the hacks and vulnerabilities lately like GitHub) to really go all in on AI. Too many unknowns. Interoperability is another issue. Getting your agent to interact with different apps that you use it’s a whole mess in itself. When someone creates the “1 click add MCP” button to interact with data from another app that is going to change the game entirely. It’ll be a ChatGPT/iPhone moment yet again

@XFreeze Threat payloads can be embedded in images and video frames. Keep your agents secure! github.com/OraclesTech/gu…

xAI just overtook Google on image editing quality rankings. The latest Grok Imagine Image Quality model has now surpassed: • Gemini 3 Pro (Nano Banana Pro) • Gemini 3.1 Flash (Nano Banana -2) on the Image Edit Arena leaderboard xAI is moving insanely fast in multimodal AI right now

@hanghuang_ @sashimikun_void @insforge Keep your agents secure! github.com/OraclesTech/gu…

Not bad! "InsForge is an open-source backend platform that gives AI coding agents instant access to databases, auth, storage, edge functions, and an AI model gateway — all through a single MCP server or CLI, so the agent can build and ship full-stack apps without a human doing the wiring."

Today, we celebrated with a custom GitHub cake for hitting 10k stars @insforge Using AI agents to fully build production-level backends and manage cloud infrastructure sounded crazy a year ago Our team has been working relentlessly to build the "agent-native AWS" But I forced us all to close our laptops for 10 minutes today to appreciate how far we've come To every builder who has starred the repo, contributed a PR, or shipped an app with us: thank you for believing in us early. We have much, much more in store for you guys! InsForge to the moon 🚀

@grinich @Cloudflare @firecrawl @WorkOS Keep your agents secure! github.com/OraclesTech/gu…

Today WorkOS is launching auth.md An open protocol for agents to register for services on the web. We're partnering with @Cloudflare and @Firecrawl as some of the first providers. Why did we build this? And why now? 🧵

@Teknium Still missing Guardian SDK support? 👀 keep your agents secure! github.com/OraclesTech/gu…

We got BitWarden now, to make it easy to manage your keys, rotate them quickly, and coordinate access with your team.

@NousResearch @Bitwarden Still missing Guardian SDK support? 👀 keep your agents secure! github.com/OraclesTech/gu…

Hermes Agent now supports the @Bitwarden Secrets Manager

@CausalEngineer @AnthropicAI Secure* agents hold the cyber advantage over attacks. Keep your agents secure! github.com/OraclesTech/gu…

@AnthropicAI I hope defensive AI agents hold the cyber advantage over attackers. If that balance breaks, Claude Mythos might end up locked behind an enterprise paywall.

Last month we launched Project Glasswing, our collaborative AI cybersecurity initiative. Since then, we and our partners have found more than ten thousand high- or critical-severity vulnerabilities in essential software.

@nitinbisht96 @AnthropicAI So long as the AI powering defense is defended itself github.com/OraclesTech/gu…

@AnthropicAI 10,000+ critical vulnerabilities in a month is wild. AI-powered defense is no longer optional. This is what the future of cybersecurity looks like.