Flare checks the libraries most AI applications actually depend on for live, high-severity advisories — plus the OWASP LLM Top 10 as a permanent reference for the risks that never show up as a CVE at all.
SyncLoading advisory data…
Your exposure
Pick the packages in your AI stack — Flare checks each one for live, high-severity advisories. See what's actually wrong with them under .
Team dashboard
Share this exact package list with your team via a link — no account or login needed. Anyone who opens the link sees the same live stack and can add or remove packages for everyone; changes sync automatically within about 20 seconds.
Anyone holding this link has full access to view and edit this shared list — the link itself is the access control, the same trust model as an unlisted document link. There's no per-person login, so don't post it somewhere public. Regenerate the link any time to cut off anyone still holding the old one. Teams untouched for 90 days are automatically deleted.
Import from a lockfile or SBOM
Upload requirements.txt, package-lock.json, pyproject.toml, or a CycloneDX/SPDX JSON SBOM — Flare selects whichever AI-stack packages below are actually in it, instead of you clicking through by hand.
Parsed entirely in your browser — the file itself is never sent anywhere. Only matches against Flare's curated AI-stack list are auto-selected below; this isn't a general dependency scanner, so the rest of your file is left alone.
Pre-selected based on the deployment pattern you told AI-SHIELD you use — adjust anytime.
Not in the list? Add your own.
Looking up a custom package sends its name to OSV.dev from your browser. Nothing else leaves this page.
Copyable digest
Loading the advisory snapshot…
Global threat map
Every package Flare watches — not just the ones in your stack above. Darker cells mean more advisories at that severity, right now. Click a package to see its advisories.
Loading the advisory snapshot…
The OWASP LLM & ASI Top 10
These are enduring risk categories, not news — they don't go stale the way a live advisory list does. Each one is flagged for whether it typically shows up as a CVE against a package you can patch, or as a property of how your application is built.
Sources
Where Flare's advisory data comes from, and what else you can add to it.
Built-in sources
OSV.dev
Aggregates GitHub Security Advisories, PyPA, npm, Go, and other ecosystem feeds into one package-precise API. This is what the ranked list above is built on — always included.
Always on
CISA Known Exploited Vulnerabilities
Fetched from CISA’s official GitHub mirror of its KEV catalog (cisa.gov itself doesn’t allow this from a browser) — this is what powers the “Actively exploited” badge. On by default since it directly affects that badge’s accuracy; turning it off falls back to a small hand-maintained list rather than an empty one.
GitHub Security Advisories
Queried directly for your selected stack, filtered by ecosystem and package name. Most of these already reach you via OSV.dev under the same GHSA ID — only advisories genuinely missing from OSV get added, so turning this on won't create duplicates.
AI Stack CVEs (assurestart.co)
A live CVE feed pre-scoped to the exact package/vendor names this page tracks (openai, langchain, litellm, huggingface, vllm, and more), high-severity only. Merges NVD, MITRE’s CVE List, CISA KEV, and CISA Vulnrichment. Shown as informational items below, not mixed into the ranked list — same as any other feed.
Package maintenance status
A CVE count of zero doesn’t mean a package is safe to keep depending on — an abandoned or maintainer-deprecated package with no CVEs yet is still a real supply-chain risk. Checks each of your selected PyPI/npm packages directly against the npm registry and PyPI’s own API for its last release date and any maintainer deprecation notice. Off by default: unlike the checks above, this is one request per selected package to two more third-party registries.
EPSS exploit probability
Severity and CISA KEV tell you how bad an advisory is and whether it’s confirmed exploited — EPSS (FIRST.org’s Exploit Prediction Scoring System) adds a continuous 0-100% estimate of the odds it gets exploited in the next 30 days, so two Critical, non-KEV advisories can still be told apart. Feeds into Top Priorities’ ranking; only ever covers advisories with a CVE. Off by default: one more third-party request beyond OSV/KEV/GHSA.
Your feeds
Add an RSS or Atom feed from anywhere else you follow. These aren’t security advisories — no severity, no CVSS, nothing to rank — so they're listed separately below, never mixed into the ranked exposure list.
Fetched directly from your browser, right now and again on every reload. Most third-party feeds don’t allow this from other websites (no CORS) and will show as unavailable rather than fail silently — nothing routes through a server of ours either way. Nothing beyond the feed URL itself is sent anywhere.
Post to Slack / Teams
Paste an incoming webhook URL to push updates straight into a channel — sent directly from your browser to Slack/Teams, no server of ours in between. Important: this only fires while this browser tab is open. Flare is a static page with nothing running in the background, so “automatic” here means “automatic for as long as you keep this tab around,” not a 24/7 monitor — leave a tab open (or use Post now, below) if you need a guaranteed alert.
Compare anonymously
Opt in to send only aggregate counts (how many advisories, what share are critical, stack size) to an anonymous benchmark on a server I control — never package names, CVE/advisory IDs, or anything else that could identify your stack. Nothing is sent unless you click the button below.
Filters
Narrow the ranked list by fix/workaround status, your watch list, or date.
Remediation SLA targets
Days from publish date to remediate, by severity. An advisory open longer than its target is flagged “SLA breached.” Set a severity to 0 to stop tracking it.