Jennifer is a general-purpose programming language, not a sandbox. A Jennifer program has full access to the host - files, network, subprocesses - exactly like a Python or Node program. That surface is a language capability, not a vulnerability: there is no trust boundary between a script and the host, so "a script can read any file / dial any address / run any program" is by design, not a bug. Do not run untrusted Jennifer code on a host you care about - isolate the whole interpreter at the OS level (a container, a VM, an unprivileged restricted user). A capability-based in-process sandbox is planned (DRAFT#11). The full model is in docs/technical/security-model.md.

A real bug is one that harms a program using the stdlib correctly, driven by untrusted data rather than untrusted code:

• Protocol / injection - a module placing a caller value on a wire without sanitizing it (CRLF injection, JWS/JSON malformation, algorithm confusion).
• DoS from untrusted input - a fatal/uncatchable crash or unbounded memory/CPU from an attacker-shaped document, message, or response.
• Cryptographic weakness - predictable randomness, a timing side channel, a verification/downgrade bypass.

Not a vulnerability (won't be treated as one): the stdlib's fs / net / os / sql / meta host access, an explicit opt-in like TLS skipVerify (verification is on by default), or a program that itself passes untrusted input to a dangerous call. See the security-model doc for the reasoning and docs/technical/rejected.md for previously-declined scanner findings.

Please report suspected vulnerabilities privately to <security@jennifer-lang.dev> rather than opening a public issue. A useful report describes (rather than patches):

• the affected file / module and a minimal reproducer,
• the untrusted-data path (who supplies the malicious input, and how it reaches the sink),
• the impact.

Jennifer is a pre-1.0, single-maintainer project: there is no security team, response SLA, or formal embargo process. Reports are read at the address above and handled as maintainer capacity allows; disclosure timing is coordinated case by case, and fixes ship with a regression test.

Please send a description, not patch code. The project does not yet have a contribution / inbound-licensing policy (the governance work is tracked as M29 in the milestones), so unsolicited code cannot be merged cleanly - the maintainer writes the fix from your report, which keeps the tree's copyright unambiguous.

Jennifer is pre-1.0 (0.x); only the latest main receives fixes.