BMP, GIF, WAV, RAW, RAW\_PADDED, CSV and PROTO formats. All other input formats,
including tensorflow-io should be sandboxed if used to process untrusted data.

For example, if an attacker were to upload a malicious video file, they could
potentially exploit a vulnerability in the TensorFlow code that handles videos,
which could allow them to execute arbitrary code on the system running
TensorFlow.

    otherwise. Do not persist these and assume they can be read by a future
    version of the library.

5.  Our classes are not designed to protect against a malicious caller. You
    should not use them for communication between trusted and untrusted code.

6.  For the mainline flavor, we test the libraries using OpenJDK 8, 11, and 17

	if ok {
		parentInClaim, ok := p.(string)
		if !ok {
			// Reject malformed/malicious requests.
			return false
		}
		// The parent claim in the session token should be equal
		// to the parent detected in the backend
		if parentInClaim != parentUser {
			return false
		}
	} else {
		// This is needed so a malicious user cannot
		// use a leaked session key of another user
		// to widen its privileges.

   *
   * <p>Note that the given byte array may be passed directly to methods on, for example, {@code
   * OutputStream} (when {@code copyTo(OutputStream)} is called on the resulting {@code
   * ByteSource}). This could allow a malicious {@code OutputStream} implementation to modify the
   * contents of the array, but provides better performance in the normal case.
   *
   * @since 15.0 (since 14.0 as {@code ByteStreams.asByteSource(byte[])}).
   */

* 📝 Reword in docs, from "have in mind" to "keep in mind". PR [#10376](https://github.com/tiangolo/fastapi/pull/10376) by [@malicious](https://github.com/malicious).
* 📝 Add External Link: Talk by Jeny Sadadia. PR [#10265](https://github.com/tiangolo/fastapi/pull/10265) by [@JenySadadia](https://github.com/JenySadadia).