Company
HackerOne is a vulnerability coordination and bug bounty platform that connects organizations with security researchers to find and report security flaws.
Rapid7 built a working command-execution exploit for GitLab's CVSS 10.0 CI/CD regex overflow, CVE-2026-97229. Rapid7 says its route needs a malicious CI config pushed to a public project first, a prerequisite that conflicts with GitLab's PR:N rating.
Reality
GitLab patched CVE-2026-97229, an integer overflow in its CI/CD regex compiler that lets unauthenticated users run code on self-managed servers. Separately, Let's Encrypt's 64-day production certificates start February 10, 2027, so renewal automation needs a test run first.
Reality
GitLab released critical patches 19.4.2, 19.3.5 and 19.2.8 on October 12 to fix a CVSS 10.0 code-execution flaw. Self-managed installations are told to upgrade immediately, while GitLab.com is already patched and Dedicated customers need do nothing.
Reality
Bloom Security counted more than 1,500 npm packages exposed because Trusted Publishing bound publish rights to a GitHub repository name. GitHub says it has shipped fixes, and Bloom's Golan Myers said it is unknown whether anyone exploited the flaw.
Perspective Coverage
3 publishersReality
GitLab fixed CVE-2026-97229, a maximum-severity flaw that could let remote attackers run code, in releases 19.2.8, 19.3.4 and 19.4.2. GitLab.com already runs the fix and Dedicated customers need do nothing, so the work falls on self-managed operators.
Perspective Coverage
7 publishersReality
A HackerOne report says DuckDuckGo's semver-label.yml ran code from any forked pull request with secrets in its environment. Any team can search its own workflows for pull_request_target to find the same pattern.
Reality
Hacktron chained a heap overflow in libheif to an OpenAI SSO flaw and opened a pull request in OpenAI's internal monorepo in under 72 hours. The bug had been fixed upstream a year earlier, but the fix never reached Debian's packages.
Reality
GitLab has fixed CVE-2026-90970, a CVSS 9.9 sandbox escape that lets an authenticated Duo Agent Platform user run commands on self-hosted AI Gateways. Anyone running one of those gateways needs build 19.2.4, 19.3.2 or 19.4.1, all released October 2, to close it.
Reality
Google stopped taking product vulnerability reports for its open-source bug bounty on October 1 after a flood of invalid AI-generated submissions. Any team that takes outside security reports faces the same imbalance, with reports now cheap to write and as costly as ever to check.
Perspective Coverage
15 publishersReality
The libheif bug was fixed upstream a year earlier without a security label or a CVE, so a Discourse image check that ignored HEIC left it reachable, and forum sign-in tokens carried full API access to the accounts behind them.
Perspective Coverage
6 publishersReality
WatchTowr reproduced CVE-2026-19478 from the advisory and patch alone, then caught the first exploitation attempts on its honeypots. Self-managed owners do not get a week to schedule this.
Perspective Coverage
5 publishersReality
HackerOne says companies cut average vulnerability fix time from 135 days to 62, yet validated findings left unresolved rose 131% over two years. That leaves time-to-fix a poor guide to exposure, and paying the debt down means taking capacity away from feature work.
Reality
Cantina released apex-flash-1, an open-weights vulnerability-research model it says solved 40 of 60 tasks for $2.38, against $74.68 for Claude Opus 5 High. There is no hosted endpoint, so teams download the 321-billion-parameter weights, pay for their own inference and verify the numbers themselves.
Reality
GitLab patched CVE-2026-90970, a sandbox escape that lets any authenticated Duo Agent Platform user run arbitrary commands on a self-hosted AI Gateway. Customers on GitLab's hosted gateway are already protected, so the upgrade falls to Self-Managed shops that run their own.
Perspective Coverage
4 publishersReality
Cloudflare fixed a Containers flaw that let paying customers read other tenants' leftover disk data, found on 18 of 24 production tries. Sandboxes, the product it sells for running untrusted and AI-written code, was affected too.
Perspective Coverage
3 publishersReality
Aikido found a dozen live GitLab 'email work item' addresses in public docs, each carrying a long-lived token that lets any mailbox act as its owner. GitLab calls it intended behavior, so maintainers must pull the address and reset the token.
Perspective Coverage
3 publishersReality
Four HackerOne reports cited in a dev.to analysis trace to S3 write scope set wider than the one object an upload needs. Only one of them is the starts-with key condition that turns a signed upload form into write access across a whole prefix.
Reality
Copilot Autofix swapped an env-var-and-jq pattern for inline interpolation in a public GitHub Actions file, and Wiz's autonomous agent exploited it inside a week.
Reality
Wiz says its Red Agent found and exploited a GitHub Actions injection that an AI "autofix" commit introduced five days earlier, reaching Snowflake's internal Jira with no human in the loop.
Perspective Coverage
4 publishersReality
Adversa AI says its Cryptographic Context Injection recovers hostile prompts inside the code sandbox, where filters do not look. xAI has not replied; Google scopes jailbreaks out entirely.
Perspective Coverage
6 publishersReality
Security · September 3, 2026 · Confirmed4 publishers
Science · September 11, 2026 · Confirmed4 publishers
Security · September 18, 2026 · Confirmed2 publishers