Security alerts
Security alerts for AI-built apps
Incidents that matter to founders running apps built with Lovable, Replit, Cursor, Bolt or Claude Code, and to the hosting, PHP and AI providers those apps depend on. Each alert says who is affected, what to do now, and where to read more. We cite primary sources and update entries as the facts change.
Last updated
Railway: customer environment variables may have been decrypted
Railway detected anomalous activity in the systems that store customer environment variables. It says an unauthorized party may have accessed encrypted variables and, for some of them, obtained the decrypted contents. It paused deployments, asked every customer to rotate all variables, and links the incident to attacks on platforms aimed at compromising AI inference providers.
Who is affected: Anyone who stores secrets in Railway environment variables: AI provider keys, Stripe keys, database URLs, cloud and private registry credentials.
What to do now
- Rotate every Railway environment variable, starting with AI provider keys, then payment, database, cloud and authentication secrets.
- Rotate at the provider first, then update the value in Railway.
- Set spending limits on new AI keys and review provider usage and audit logs for unfamiliar activity.
Read more on Zenveus
- Railway breach: rotate every secret, starting with AI keys →
- AI API key exposed in frontend code →
- Database URL exposed in the browser →
Sources: Railway: security disclosure · TechCrunch · GitGuardian
PHP2Shell: two critical PHP flaws (CVE-2026-109618, CVE-2026-109619)
PHP fixed two flaws rated 9.2 that can allow remote code execution without a login, in PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. Cloudflare reports millions of exploit attempts since disclosure, and Amazon saw an attempt on October 17. Agencies in Germany, the Netherlands and Japan reported no confirmed break-ins as of October 22, and several organizations took websites offline as a precaution.
Who is affected: Any server where PHP answers web requests: WordPress and other PHP CMS sites, Laravel or Symfony APIs, admin tools and hosting panels, on any web server.
What to do now
- Check the PHP version your web server runs (not only the CLI) and update to a patched release, or ask your host for a date.
- If you can't patch yet, apply PHP's mitigations at the server level and consider taking the site offline.
- Look for signs of compromise from October 17 onward.
Read more on Zenveus
- PHP2Shell: what to check on your PHP servers today →
- Your host hasn't patched PHP yet: what you can do today →
- How to tell if your WordPress site was hacked →
Sources: PHP: security release · GHSA-w8hw-w9fx-j79w · GHSA-6wmf-4rv5-p3rp · The Next Web
OpenCode Console: linked AI provider keys exposed
OpenCode said a large-scale distillation attack affected 1.4% of OpenCode Console users, those who had linked external providers such as Anthropic, OpenAI and Google. Its email to users gives October 17, 00:30 UTC as the earliest possible exposure. Some users reported unauthorized charges on their AI accounts.
Who is affected: OpenCode Console users who saved third-party AI provider keys in their workspace.
What to do now
- Revoke and replace any AI provider key you linked to the OpenCode Console.
- Review usage on those provider accounts from October 17 onward and contact the provider about unexpected charges.
- Use a separate key per tool, with a spending limit, so one exposure is contained.
Read more on Zenveus
Sources: BleepingComputer · TechCrunch
How are alerts chosen and kept up to date?
What gets listed. Incidents that affect the hosting, runtimes, AI providers and developer tools that AI-built apps commonly depend on, where there is something specific a founder or developer should do.
How we source it. From the vendor's own disclosure or advisory first, then established security reporting. Unconfirmed claims are left out until a primary source confirms them.
What the status means. Active: act now. Monitoring: the main actions are known and the vendor is still investigating. Resolved: kept for reference.
Frequently asked questions
How do I know whether an alert affects my app?
Check the "Who is affected" line against your stack: where the app is hosted, which runtime serves it, and which AI and payment providers hold keys for it. If you are not sure what your app depends on, start with an inventory of every service and secret it uses.
What should I do first when a provider I use is breached?
Rotate the secrets that provider held, starting with keys that can spend money, then keys that reach customer data. Then review each provider's usage and audit logs for activity you don't recognise.
Can Zenveus check my app for me?
Yes. The 48-Hour Verdict has a named senior engineer review your codebase against the Zenveus Production Readiness Standard for $299. For a deeper review of who can reach your data and secrets, there is the Security & Data Isolation Audit.
Want an engineer to check your app?
A fixed-scope review of authentication, authorization, tenant isolation, secrets, APIs, webhooks, dependencies and storage permissions.
See the audit →A named senior engineer reviews your codebase against the Production Readiness Standard and signs a written decision with evidence and next steps.
Get the Verdict →