Skip to content

Company

OX Security

OX Security is a cybersecurity firm focused on software supply chain security, researching vulnerabilities and malicious packages across open-source ecosystems.

Known aliases

  • OX
  • Ox Security
  • OX Security Research

Current stories

buildOne report1 publisher

DeepSeek Harness lets a sandboxed agent escalate itself to full access via an unauthenticated API

DeepSeek Harness, DeepSeek's open-source agent runner, carries a 9.4-severity flaw that lets a sandboxed agent grant itself full access with one command. Its control API has no authentication and trusts a client-supplied Host header.

Publishers:dev.to

Reality

Evidence45
Adoption55
Hype gap+20
Incentives
Not scored
Confidence40
securityOne report1 publisher

OX Security scan of 15,465 MCP servers puts vetting on the companies installing them

OX Security scanned 15,465 public MCP servers in five registries and found no marketplace review or guardrails. Six of the listed servers sit on expired domains anyone can register for $4 to $12 a year, and a buyer would inherit the agents still calling them.

Reality

Evidence40
Adoption45
Hype gap+30
Incentives80
Confidence50
buildOne report1 publisher

NVIDIA's Sentry enforces agent limits from a separate BlueField-4 card

NVIDIA's Open Agent Safety Platform runs Sentry, a watchdog on separate BlueField-4 cards that isolates an agent within milliseconds of crossing its boundary. For teams running coding agents, the design takes enforcement out of the agent's own process, where prompts and permission lists sit today.

Publishers:dev.to

Reality

Evidence45
Adoption
Not scored
Hype gap+25
Incentives55
Confidence50
securityOne report1 publisher

Ox Security finds nearly 16% of public MCP server hostnames resolve outside the US

Ox Security found nearly 16% of 5,095 hostnames listed in three public MCP registries resolve outside the US, including to Russia and China. MCP has no notion of region, so cloud residency rules end where a company's AI agents connect out.

Reality

Evidence40
Adoption35
Hype gap+25
Incentives65
Confidence45
buildOne report1 publisher

About 500 poisoned documents backdoored models at both 600M and 13B parameters

A dev.to survey walks seven AI supply chain entry points and the named incidents behind each. The two dataset-poisoning numbers in it are the ones that should change how a model review is scoped.

Publishers:dev.to

Reality

Evidence55
Adoption45
Hype gap−10
Incentives30
Confidence52
productOne report1 publisher

A spoofed Host header let DeepSeek Harness agents switch off their own sandbox

OX Security found that DeepSeek's coding harness would let an agent unlock itself simply by asking, because the gate trusted a name the caller supplies. That makes agent isolation a claim you have to attack before you report it as a control.

Publishers:devops.com

Reality

Evidence70
Adoption55
Hype gap+18
Incentives65
Confidence60
buildOne report1 publisher

Google's CDN delivered version 1.7.3.0 of an extension it had pulled six months earlier

Netskope Threat Labs puts the update between 20 and 31 July, and describes a Chrome sidebar that still earned an affiliate commission as users removed it. That says more about the update channel than about the install.

Publishers:dev.to

Reality

Evidence44
Adoption52
Hype gap+12
Incentives55
Confidence46
securityOne report1 publisher

A poisoned Nx Console build rode VS Code's auto-update into GitHub's own repositories

CISA says a trojanized extension version, 18.95.0, reached a GitHub employee's machine without anyone installing it, and internal repositories left from there. CVE-2026-48027 is now in the KEV catalog.

Publishers:cisa.gov

Reality

Evidence76
Adoption58
Hype gap−8
Incentives24
Confidence74
buildOne report1 publisher

Your scanner finds it in seconds; the average fix now takes 252 days

GitHub now sells Advanced Security as two SKUs, at $19 and $30 per active committer per month. Neither one routes a finding to an owner or enforces a deadline.

Publishers:dev.to

Reality

Evidence28
Adoption
Not scored
Hype gap+34
Incentives74
Confidence33
buildOne report1 publisher

Amazon Q executed code from any repo you opened, and it is not the only one

Wiz says the VS Code extension auto-loaded MCP server configs from the workspace and handed spawned processes the developer's full environment. Fixed in language server 1.65.0.

Publishers:wiz.io

Reality

Evidence62
Adoption30
Hype gap+15
Incentives70
Confidence55
scienceOne report1 publisher

OX Security says MCP command execution is a design choice, so server owners own the risk

The vendor reports 10-plus CVEs and up to 200,000 exposed instances, and says Anthropic declined to change the protocol, describing the behaviour as expected.

Publishers:ox.security

Reality

Evidence32
Adoption34
Hype gap+46
Incentives86
Confidence33
securityOne report1 publisher

Agent skills load at runtime with no signing, no provenance, and an 82% marketplace pass rate

Nine of eleven MCP marketplaces accepted proof-of-concept malware with zero review, and a fake agent skill cleared both Cisco's and NVIDIA's scanners to reach roughly 26,000 corporate agents.

Publishers:scworld.com

Reality

Evidence24
Adoption38
Hype gap+26
Incentives62
Confidence30