Skip to content
Menu
Roman Haltinner

By prioritizing vulnerabilities based on actual business impact and providing validated, actionable remediation, 0.security enables security teams to focus their resources where they matter most. A compelling AI-powered security solution for enterprise environments.

Roman Haltinner

Partner & Europe West Cybersecurity Competency Leader, Switzerland, at EY

Powerful enough to secure governments, hack operating systems & fix software billions rely on.

3+ billion
devices run software that we have secured
Cyber models
  • Anthropic Cyber program
  • OpenAI Daybreak
  • Microsoft Security Advisor

We shipped security fixes for

  • AMD
  • Qualcomm
  • Cisco
  • Arm
  • Google
  • Meta
  • IBM
  • NVIDIA
  • Intel
  • Red Hat
  • Ericsson
  • NXP
  • SUSE
  • Citrix
  • Alibaba

Research findings

CriticalRX buffer overflow on zero-length serial framesLinux kernel · mctp
CriticalArbitrary code execution via crafted expressionsjsonata
CriticalCross-tenant agent API-token minting@paperclipai/server
MediumAnonymous /api/systemstatus leaks exception detailSwiss Federal Chancellery · government code
HighSlab use-after-free in AEAD decrypt completionLinux kernel · TIPC
HighUse-after-free in MACsec offload RXLinux kernel · mlx5e
HighMissing CAP_NET_ADMIN check on changelinkLinux kernel · vxlan / geneve
HighOOB read on frames shorter than the auth tagLinux kernel · mac802154
HighMemory corruption in 802.15.4 llsec decryptLinux kernel · mac802154
HighUse-after-free of metadata_dst in eswitch repr releaseLinux kernel · ice
HighResource-exhaustion denial of servicejsonata
HighUse-after-free in LE Audio CIG/CIS setupLinux kernel · Bluetooth
HighStack buffer overflow in NFC digitalLinux kernel · NFC
HighDouble-completion / double-free on resumeLinux kernel · xen-blkfront
HighCertificate-chain validation bypassnode-forge
HighDenial of service via unbounded recursionprotobuf.js
HighInteger overflow in HTTP chunked-transfer parser → heap OOB readnng (nanomsg)
HighOOB from unvalidated ioctl buffer sizesLinux kernel · atomisp
HighUse-after-free on inrange_timer at teardownLinux kernel · HID uclogic
HighStale pen_input pointer on partial registrationLinux kernel · HID uclogic
HighUse-after-free of in-use VP9 framesLinux kernel · meson vdec
MediumOOB read on LLCP PDUs shorter than the headerLinux kernel · NFC
CVSS pendingNULL-deref DoS paths via AMDXDNA_EXEC_CMDLinux kernel · amdxdna
CVSS pendingOOB read from assoc length underflowLinux kernel · ath6kl
CVSS pendingNULL deref on HT-cap without HT-oper in TDLSLinux kernel · mwifiex
CVSS pendingOut-of-bounds read from unset MAC header on packet-socket TXLinux kernel · net/packet
MediumOOB read in LLCP connect_sn TLV walkLinux kernel · NFC
MediumOOB read in st21nfca ATR_REQ handlingLinux kernel · NFC
CVSS pendingOOB read from unvalidated directory-index countsLinux kernel · ocfs2
CVSS pendingUse-after-free from carrier_work rearm on disconnectLinux kernel · ipheth
CVSS pendingUse-after-free of ghl_poke_timer / ghl_urb at driver unbindLinux kernel · HID sony
MediumHeap out-of-bounds write in ciscodump extcapWireshark
MediumTemplate injection → arbitrary file readUptime Kuma
MediumStack overflow via deeply nested collectionsyaml
MediumOut-of-bounds read in LLCP SNL TLV parserLinux kernel · NFC
MediumInteger overflow in frame-size calculationLinux kernel · atomisp
MediumOOB read from unvalidated AV1 frame indicesLinux kernel · v4l2-ctrls
MediumOOB in NPU cmdstream tile validationLinux kernel · accel/ethosu
MediumOOB read in firmware-request handlerLinux kernel · Bluetooth btnxpuart
MediumOOB read in HT/VHT capability IE parsingLinux kernel · mwifiex
MediumOOB read in pairwise-cipher OUI walkLinux kernel · mwifiex
MediumECRED deferred-recvmsg race → list corruptionLinux kernel · Bluetooth L2CAP
MediumOOB read in port100 frame length handlingLinux kernel · NFC
LowReachable WARN DoS in rsa-pkcs1pad empty digestLinux kernel · crypto

A sample of what we've found. More are disclosed continuously, and others stay under embargo and NDA.

  • Persistent

    Agents keep watching, on a schedule and on every pull request.

  • Any model

    Your provider key or a local model.No vendor lock-in.

  • Runs anywhere

    Locally, in your cloud or air-gapped.In your own environment.

Build your own army of security agents.

Or start from ready-made ones. They find, prove, fix and verify, then keep watching your code, dependencies, infrastructure and AI agents.

  1. Find vulnerabilities

    Review source code, dependencies, packages and live web targets. Pick a module, or let 0 choose one.

  2. Prove and fix

    Show each bug with a probe, write the patch and verify it against your tests before you apply it.

  3. Reverse engineer binaries

    Analyze binaries with Ghidra, and run kernel and memory-safety fuzzing alongside source review.

  4. Test AI agents

    Run prompt-injection campaigns against LLM apps, and check whether an agent took a prohibited action.

  5. Keep watching

    Run agents on a schedule, on every pull request or in CI, with JSON and SARIF reports.

  6. Build your own

    Give an agent a scope, skills, plugins and memory, or start from a ready-made one like the dependency watcher.

Sovereign by design.

  • Locally

    On your machine, in CI or in a container, with any model provider or a local model.

  • Your cloud

    Inside your own VPC, with the models you already run there.

  • Air-gapped

    On your own hardware with local models, with no outbound connection.

Enterprise security automation

Run it across your organization, scoped with your team.

Explore enterprise

Questions about 0security

What is 0security?

0security is software infrastructure to deploy sovereign, persistent AI security agents that find and fix security risks across any tech stack, work with any model, and run anywhere.

Is 0security open source?

Yes. The code is public on GitHub, licensed under MIT or Apache-2.0, at your choice.

Can I run it on my own infrastructure?

Yes. It runs on your machine, in your CI or in a container, with your own model keys or a local OpenAI-compatible server. Only the context you send to your chosen model leaves it.

Which models does it support?

Connect your own provider, such as Anthropic, OpenAI, Azure OpenAI, OpenRouter, DeepSeek or xAI, or any OpenAI-compatible server, including a local one. Agent workflows need a model that supports tool calling.

What can it scan?

Source code, packages on npm, PyPI, Cargo and OCI, web apps and APIs, MCP endpoints, binaries, cloud and identity posture, and AI agents. For dependencies, 0 deps scan and 0 deps fix cover npm, pnpm, Cargo and PyPI.

How does it prove and fix a vulnerability?

Scans check each finding before they report it, and 0 scan repair writes a probe that shows the bug. A patch counts as verified only when the probe and your tests pass on a fresh patched checkout. It never merges or deploys; you review every change.

What are persistent security agents?

They are agents with a scope, an autonomy level, a daily budget and memory. They run on a schedule, on GitHub pull requests or by hand. Each one starts paused and needs your review before it runs on its own.

Does it work in CI?

Yes. The GitHub Action runs the open-source CLI, keeps JSON and SARIF reports and can fail the build on findings. On GitLab, Jenkins and other runners, call the CLI directly.

Is it safe to point at my systems?

Only test systems you own or are authorized to test. Turn on the scope plugin to limit live testing to the hosts in your scope file. The default shell runs on your host, so use an isolated environment for untrusted code and targets.

How do I get started?

Install it with curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash, then run 0 and connect a model provider. To roll it out across a team, book a demo.

Put security agents to work.

Start with the open-source code, or talk to us about running it across your organization.

curl -fsSL https://raw.githubusercontent.com/0sec-labs/0security/main/install.sh | bash