A skill file. A rules file. An MCP server configuration. Three lines of plain text, commitzoo li cov ntaub ntawv sau tseg, tshuaj xyuas zoo li cov ntaub ntawv sau tseg, thiab tsis muaj ib qho zoo li cov lej. Txawm li cas los xij, txhua tus tuaj yeem rov sau dua yam uas koj tus pabcuam AI raug qhia kom ua thiab yam uas nws raug tso cai ua. Qhov ntawd yog qhov tseeb tsis xis nyob tom qab kev ruaj ntseg AI. nyob rau hauv 2026Cov lag luam siv ob xyoos los txhawj txog cov lej uas AI tsim muaj. Qhov teeb meem nyuaj dua yog cov khoom siv AI nws tus kheej: cov qauv, cov neeg sawv cev, MCP servers, thiab cov ntaub ntawv teeb tsa uas tam sim no zaum nrog koj cov lej qhov chaw thiab qhov chaw qhib, feem ntau tsis tau sau thiab tsis tau tshuaj xyuas. Qhov no yog vim li cas kev ruaj ntseg ntawm cov khoom siv AI tau dhau los ua nws tus kheej txoj kev qhuab qhia, thiab vim li cas kev xaiv lub tuam txhab kev ruaj ntseg AI zoo tseem ceeb npaum li kev xaiv lub scanner zoo.
Qhov chaw tawm tsam uas tsis muaj leej twg npaj nyiaj rau
Software used to have a handful of places an attacker could land: the code, the dependencies, the pipeline. AI tau ntxiv ob qho ntxiv, thiab ob qho tib si pub ncaj qha rau hauv AI cov khoom xa tuaj.
The model and the agent. Kev lom cov cuab yeej, kev txhaj tshuaj sai sai, kev ywj pheej ntawm tus neeg sawv cev uas mus deb dua li leej twg xav tau. Cov lus qhia zais cia hauv cov lus piav qhia ntawm MCP server tuaj yeem hloov pauv qhov uas tus copilot ua, thiab tus tsim tawm yeej tsis pom nws tshwm sim.
Tus tsim tawm tus kheej ib puag ncig. IDEs, AI copilots, MCP servers, agent CLIs. Invisible to legacy AppSec scanners, which don’t know what a model is, and invisible to EDR, which watches the operating system and has no idea what a dependency or an MCP call is.
Tsis muaj ib qho ntawm qhov no yog kev xav xwb. Hauv kaum yim lub hlis dhau los:
- Ib qho "qhov rooj zais cia" uas zais cov ntaub ntawv cai Unicode cia cov neeg tawm tsam txhaj cov lus qhia tsis pom rau hauv cov ntaub ntawv teeb tsa Copilot thiab Cursor nyeem, ntsiag to zais cov lej uas tus pabcuam tsim. GitHub tau ntxiv ib qho lus ceeb toom rau nws hauv xyoo 2025.
- Ib qho teeb meem hais kom ua hauv lub choj MCP nrov (CVSS 9.6) tau ntaus ntau dua 400,000 rub tawm ua ntej nws tau kho, thawj qhov xwm txheej sau tseg ntawm kev ua tiav tag nrho cov lej chaw taws teeb uas tsuas yog tshwm sim los ntawm kev txuas mus rau lub server MCP uas tsis ntseeg siab.
- Ib tug kab mob npm uas kis tau tus kheej tau ua rau cov neeg tsim khoom lawv tus kheej mus rau hauv lub tshuab xa khoom, thiab tus qauv rov ua dua ntawm qhov ntsuas hauv cov hlis tom ntej thoob plaws lwm lub ecosystem, ib phau ntawv AI kev ruaj ntseg tsis ua haujlwm.
- Cov kws tshawb nrhiav pom tias feem ntau ntawm cov pob khoom uas LLMs pom zoo tsis muaj nyob txhua, "slopsquatted" sau npe tus neeg tawm tsam ua ntej tus tsim tawm tiag tiag puas tau thov tus qauv kom import lawv.
Google txoj kev tshawb fawb txog kev ruaj ntseg ntawm AI software supply chain tau mus txog qhov xaus zoo sib xws los ntawm lub kaum sab xis sib txawv: cov qauv pom muaj nyob rau hauv xyoo 2023 thiab 2024 zoo li raug cai thaum nqa cov lej uas tuaj yeem rho tawm cov ntaub ntawv lossis cog lub qhov rooj tom qab thaum rub tawm, thiab qhov kev kho tsis yog ib pawg cuab yeej tshiab ntau npaum li kev siv cov kev qhuab qhia ntawm cov khoom xa tuaj, xws li qhov chaw thiab kev kos npe, rau cov khoom cuav uas tsis muaj leej twg tau taug qab ua ntej. Qhov ntawd yog qhov teeb meem kev ruaj ntseg ntawm AI supply chain hauv ib kab lus: cov khoom cuav yog qhov tshiab, tab sis qhov kev qhuab qhia uas lawv xav tau tsis yog.
Why your existing tools stop short
SAST nyeem code. SCA nyeem ib daim ntawv qhia txog kev vam khom. Tsis muaj leej twg paub tias tus qauv yog dab tsi, MCP server qhia dab tsi, lossis cov ntaub ntawv txuj ci qhia tus neeg sawv cev ua dab tsi. Qhov sib txawv ntawd yog qhov chaw uas AI-era tawm tsam, nyob rau hauv qhov chaw ntawm "code peb scan" thiab "AI peb ntsiag to txais yuav."
Qhov tshwm sim yog ib pawg ntawm cov duab ntxoov ntxoo AI tsis muaj CISTam sim no peb teb tau rau: cov qauv twg peb khiav, cov neeg sawv cev twg tuaj yeem ncav cuag dab tsi, thiab MCP server twg uas ib tug neeg tau txuas rau hnub Tuesday dhau los yam tsis qhia leej twg. Kev teb cov lus nug ntawd zoo yog txoj haujlwm ntawm AI kev ruaj ntseg ntawm cov khoom siv, thiab nws yog qhov laj thawj uas cov cuab yeej AppSec feem ntau tsis ua haujlwm ntawm no.
Lub ntsiab lus ntawm kev ruaj ntseg AI tiag tiag yog dab tsi
Xygeni yog lub tuam txhab kev ruaj ntseg AI uas kho qhov no ua peb qhov kev txav mus los sib txuas thoob plaws SDLC: nrhiav pom, ntes tau, thiab siv.
Discover: know what AI you actually have
Kev tshawb pom tsis tu ncua, tsis siv neeg thoob plaws koj cov chaw khaws ntaub ntawv qhia txog txhua yam khoom muaj nqis AI: cov qauv, cov qauv, cov ntaub ntawv teeb tsa, cov ntsiab lus xaus, cov neeg sawv cev, MCP servers, cov txuj ci, cov lus qhia, guardrails, and the AI coding tools your developers are actually using. No surveys. No self-reporting. If it left a trace in a repository, it shows up in the inventory, the first and most basic requirement of real AI supply chain security.
Daim duab AI ces qhia txog seb cov cuab tam ntawd txuas li cas: tus qauv twg ib qho dataset pub, tus neeg sawv cev twg siv cov cuab yeej twg, MCP server twg nyob tom qab tus pabcuam twg. Ib qho cuab tam nyob ib leeg qhia koj me ntsis. Daim duab qhia koj qhov chaw uas kev pheej hmoo tsom mus rau.
Los ntawm qhov kev tshawb pom ntawd, Xygeni tsim ib qho AI-BOM: an audit-ready, machine-readable inventory of everything AI-related in your software. When a regulator, an auditor, or a customer asks what AI you’re running, the answer becomes a download instead of a three-week scramble.
Tshawb nrhiav: cov kev pheej hmoo uas cov tshuab luam theej ib txwm tsis tuaj yeem pom
Ib lub tshuab ntsuas AI tshwj xeeb nrhiav cov hom kev ua tsis tiav uas tshwj xeeb rau AI systems: kev txhaj tshuaj sai, kev txhaj tshuaj cuab yeej thiab kev thov cuab yeej tsis ntseeg siab, cov ntaub ntawv xau los ntawm kev nrhiav, kev hla dhau qhov kev ceeb toom ntawm lub system, kev ua haujlwm ntau dhau. Txhua qhov kev tshawb pom qhia txog OWASP Sab saum toj 10 rau LLM Daim Ntawv Thov thiab taw tes rau cov ntaub ntawv thiab kab tseeb uas tsim qhov kev nthuav tawm, tsis yog qhov ceeb toom "txheeb xyuas koj qhov kev siv AI".
The same detection layer treats skill files, rules files, and MCP configurations as the security artifacts they are, not as harmless documentation. It flags malicious or poisoned skills, inspects MCP server configurations for tool poisoning, and surfaces the prompts actually driving your AI workloads.
Tsom ntsoov rau qhov tseem ceeb: lub funnel uas txiav suab nrov, tsis yog cov ces kaum
Every finding gets filtered progressively: down to what’s reachable in application code, then to what’s genuinely exploitable, then to what sits in code your team is actively developing. What reaches a developer’s queue is the short list that actually threatens production, with the framework reference, the exposure window, and mitigation guidance attached.
Yuam kom ua: nres nws ua ntej nws khiav
Shield coj txoj cai tswjfwm mus rau tus tsim tawm qhov kawg: nws thaiv cov kev teeb tsa tsis raug cai thiab phem, cov qauv tsis tau pom zoo, thiab cov servers MCP tsis tau kev pom zoo ua ntej txhua yam ua tiav. Hauv qab nws zaum Xygeni's Kev Ceeb Toom Thaum Ntxov Txog Malware (MEW), uas ntes tau cov pob khoom phem ua ntej kos npe muaj, cov cuab yeej raws li lub koob npe nrov ntawm txheej tseem ntseeg vim tsis muaj leej twg tau tshaj tawm cov pob khoom tseem. Nws yog ib nrab ntawm kev siv AI kev ruaj ntseg: kev tshawb pom thiab kev ntes qhia koj tias muaj dab tsi tsis ncaj ncees lawm, Shield yog qhov uas ua rau nws nres.
Koj qhov kev nthuav tawm AI tsis yog tsuas yog nyob rau hauv koj tus lej AI xwb
Ib daim duab kev ruaj ntseg ntawm AI cov khoom xa tuaj xav tau ntau dua li cov qauv khoom, thiab nws tsis tshua muaj cov khoom ci ntsa iab xwb:
- Cov ntaub ntawv pov thawj ntawm tus neeg muab kev pabcuam AI sab laug hauv cov ntaub ntawv qhia, tus neeg sawv cev teeb tsa, lossis pipeline cov cav ntoo yog cov lus zais zoo li lwm yam, thiab Xygeni qhov kev tshawb pom zais cia ntes tau lawv ua ntej lawv mus txog rau hauv kev sau npe rau pej xeem.
- Kev vam khom AI thiab ML uas tsis muaj zog nqa cov CVEs ib txwm muaj, uas tau tshwm sim los ntawm kev tshuaj xyuas cov software composition uas twb tau npog tag nrho koj cov stack. Kev tshawb fawb sab nraud ntawm kev siv AI tau taw qhia tias ntau npaum li cas ntawm cov AI stack niaj hnub no yog cov pob khoom sab nraud thiab cov khoom zais, uas yog qhov kev tshuaj xyuas cov software saum npoo av tau tsim los npog.
- Cov pob khoom phem luam tawm sai dua li lwm cov lus qhia pipeline tuaj yeem teev lawv cov ntaub ntawv raug ntes ua ntej kos npe, tib lub peev xwm MEW tiv thaiv koj cov khoom xa tuaj.
Cov txheej txheem agent: DevAI thiab CoreAI
Kev tshawb pom thiab kev tshawb pom npog txog yam uas twb muaj nyob hauv koj lub chaw cia khoom lawm. DevAI works where the risk is created: inside the IDE, as a continuous, proactive layer that scans human and AI-generated code as it’s written, no prompts required. It explains the full exploit path behind a finding and proposes MCP-validated fixes a developer can apply with confidence, without breaking the build.
CoreAI nyob saum cov scanners ib leeg ua txheej txawj ntse: nws sib txuas cov code, kev vam khom, pipeline, thiab muab cov ntaub ntawv tso rau hauv ib qho qauv kev pheej hmoo, teb cov lus nug hauv hom lus ntuj, thiab tsim cov lus ceeb toom npaj rau cov thawj coj uas tus thawj coj kev ruaj ntseg xav tau los qhia tias kev tswj hwm tau tshwm sim tiag tiag, tsis yog tsuas yog thov xwb.
Txuas ntxiv qhov koj muaj hauv AI Security. Tsis txhob rhuav tshem dab tsi
Qhov kev tawm tsam feem ntau rau pawg kev ruaj ntseg tshiab yog "peb twb muaj cov cuab yeej txaus lawm." Ua ib lub tuam txhab kev ruaj ntseg AI, Xygeni tsis thov kom koj hloov dab tsi: tib qho kev faib tawm, kev piav qhia, thiab kev txiav txim siab tseem ceeb uas siv rau nws cov kev tshawb pom khiav sib npaug ntawm cov kev tshawb pom los ntawm koj cov uas twb muaj lawm. SAST, SCA, thiab cov scanners sab nraud. Koj lub stack tam sim no dhau los ua ib qho input, tsis yog ib qho kev raug mob, thiab koj txoj kev ruaj ntseg ntawm AI supply chain zoo dua yam tsis muaj qhov project rip-and-replace txuas nrog nws.
Vim li cas qhov no tseem ceeb tam sim no, tsis yog tom qab
Cov neeg tswj hwm tab tom sib sau ua ke ntawm tib qhov kev cia siab los ntawm ntau qhov kev taw qhia: EU AI Act, NIS2, thiab Spain's ENS txhua tus thawb mus rau kev sau cov khoom thiab kev taug qab rau AI systems, tib cov pov thawj uas AI-BOM tau tsim los tsim. Kev taw qhia ntawm kev mus ncig yog qhov tseeb txawm tias qhov twg cov txheej txheem ua raws cai tseem nyob ruaj khov: koj tsis tuaj yeem ua pov thawj rau AI uas koj tsis tau sau cov khoom, thiab koj tsis tuaj yeem thov kev ruaj ntseg ntawm AI cov khoom xa tuaj yog tias cov khoom xa tuaj nws tus kheej tsis pom rau koj.
Choosing an AI security company
Tsis yog txhua lub tuam txhab kev ruaj ntseg AI kos nws ciam teb rau tib qho chaw. Qee qhov nres ntawm kev luam theej koj tus kheej cov lej tsim los ntawm AI. Lwm tus nres ntawm qhov kawg. Cov lus nug kev ruaj ntseg ntawm AI cov khoom siv yog loj dua li ob daim ntawm nws tus kheej: nws npog tus qauv, tus neeg sawv cev, MCP server, cov ntaub ntawv txuj ci, thiab qhov kev vam khom ib txwm nyob hauv qab nws tag nrho. Qhov kev pom tag nrho lub neej, los ntawm kev tshawb pom mus rau kev siv, hauv ib lub console nrog rau lwm cov kev tshawb pom AppSec, yog qhov yuav tsum nrhiav thaum koj ntsuam xyuas lub tuam txhab kev ruaj ntseg AI es tsis yog ib qho cuab yeej ib qho.
Cov ntaub ntawv uas tsis muaj leej twg tshuaj xyuas tau dhau los ua txoj hauv kev. Kev ruaj ntseg ntawm AI yog qhov kev qhuab qhia ntawm kev tshuaj xyuas lawv, thiab kev ruaj ntseg ntawm cov khoom siv AI yog qhov ua rau qhov kev qhuab qhia ntawd tuav tau txij thaum kawg, ntawm tib lub platform uas koj twb tshuaj xyuas txhua yam lawm.
Saib seb koj tus AI raug tso cai ua dab tsi. Pib dawb or teem caij rau demo.
FAQ
Does Xygeni’s code ever leave my infrastructure?
No. Scans run inside your own environment, and source code is never uploaded to Xygeni’s servers. The AI inventory and AI-BOM are built from what the scanner sees locally, not from a copy sent externally.
Qhov txawv ntawm AI Security, DevAI, thiab CoreAI yog dab tsi?
Kev Ruaj Ntseg AI tshawb pom thiab ntes tau: nws tsim cov khoom lag luam AI, AI-BOM, thiab nrhiav kev pheej hmoo xws li kev txhaj tshuaj sai lossis cov ntaub ntawv txuj ci lom. DevAI ua haujlwm hauv IDE thaum cov neeg tsim khoom sau cov lej, tawm tswv yim kho thaum lawv mus. CoreAI zaum saum toj ob qho tib si, sib txuas cov kev tshawb pom thoob plaws hauv lub platform thiab teb cov lus nug txog koj txoj kev ruaj ntseg hauv cov lus ntuj.
Xygeni ua haujlwm nrog AI kev ruaj ntseg twg?
Cov kev tshawb pom qhia txog OWASP Top 10 rau LLM Applications, OWASP Top 10 rau MCP, thiab OWASP Top 10 rau Agentic Skills, nrog rau NIST SP 800-218A thiab CISCov lus qhia ntawm A/G7 txog cov nqi khoom siv AI. Daim ntawv qhia ntawd yog qhov ua rau AI-BOM siv tau ua pov thawj ua raws li kev cai es tsis yog tsuas yog cov khoom muag xwb.
Qhov no puas yuav ua rau txhua lub tsev qiv ntawv AI lossis qauv muaj kev pheej hmoo?
Tsis yog. Lub funnel prioritization ua rau cov kev tshawb pom nqaim mus rau qhov uas ncav cuag tau hauv cov lej thov, siv tau tiag tiag, thiab nyob rau hauv kev tsim kho nquag, yog li daim ntawv teev npe uas tus tsim tawm pom luv luv, tsis yog pov tseg ntawm txhua yam khoom siv AI pom.
