Over the past few weeks, supply chain attacks have hit the open-source packages almost every web product is built from, one after another, alongside critical flaws in widely used web frameworks that could let an attacker run their own code on a company's servers. Gydr was not breached, and nothing we know of was aimed at us. But the businesses on Gydr trust us with their customers' conversations, their catalogues and their accounts, so we treated the wave as a deadline rather than a headline.

What changed in the software we depend on

Every modern product stands on thousands of packages written by other people. A supply chain attack poisons one of them and waits for it to be installed. So we went through everything Gydr depends on.

  • Every critical advisory is closed. We resolved all critical and almost all other published security advisories in the components Gydr is built on, and moved to the latest security releases of our runtime and build tools.
  • New code waits before we trust it. A freshly published version of anything we depend on is not accepted until it has been public for a day, the window in which most poisoned releases are caught and pulled.
  • A package cannot quietly get less trustworthy. If a dependency's publishing record becomes weaker than it was, for example a release that is no longer signed by its usual build, it is refused.
  • Installing is not running. Only a short, reviewed list of packages may run code when they are installed.

What changed in Gydr itself

The same weeks were a good time to close doors in our own product, too.

  • Your AI agent cannot change who can sign in. An AI agent connected to your account can prepare a change to how your visitors sign in, but nothing happens until you approve it in your own browser, after checking every value.
  • Secrets never pass through a chat. A password, an API key or a test user's token is typed on a Gydr page you open yourself, never in the conversation with an agent, and a test token is deleted after an hour.
  • A suspended account is fully paused. Its chat widget, messaging channels, APIs and AI agents all refuse it, it spends nothing, and it sends no email.
  • Notifications go only to people who can act on them. A billing or credit alert reaches the members who can change billing, not everyone who can see it, and an unsubscribe link carries no name or account.
  • Addresses you give us cannot reach inside. When Gydr calls an address a business provides, such as a webhook or an API, the request is refused if that address points into a private network.

What we do not claim

No system is unbreakable, and we will not pretend ours is. Three advisories remain open, two of them in a library we use to read uploaded spreadsheets, whose fix is not yet published through the usual channels. We track every one of them until it is closed. What we can promise is the habit: when the industry is attacked, we check our own house first, and we tell you what we changed.

Frequently asked questions

Was Gydr affected by the recent supply chain attacks?

No. Gydr was not breached, and we found no sign of an attack on our systems. The changes in this release are precautions, made because the risk to every web product went up.

Do I need to do anything to stay protected?

No. Every change here is on our side. If your team uses an AI agent with your Gydr account, you now approve sign-in changes in your browser, which is the point.

What happens to my visitors' data when an account is suspended?

Nothing runs on it. A suspended account answers no visitors, spends no credits and sends no emails until it is reactivated, and its data stays where it is.

How do I report a security issue to Gydr?

Write to security@gydr.ai, the address in the security.txt file on gydr.ai, which is where we take security reports.

The full list of changes is in the 4.3.0 release notes.