anthropic 1.17.0

Source: https://github.com/anthropics/anthropic-sdk-python
Tag scanned: v1.17.0
Compared with: v1.16.0 (0 signals introduced since)
Files: 2700 scanned, 32 skipped (binary, oversized, or unrecognised)
Signals: 0 critical · 0 warn · 5 info
Scan time: 7.3 s
LLM review: Reviewed by claude-haiku-5-5 (Anthropic): 5 of 5 warn and critical signals; 5 demoted to info.
Disclude version: 4.1.0
Scanned (UTC): 2026-10-21 21:14

Signals

Ordered by severity, then signals introduced in this release. Hidden characters are shown as ⟨U+…⟩.

info ~~warn~~ → info by review dynamic-execution tests/lib/test_credentials.py:1902:9

`exec` called on a non-literal expression (ast pass, confidence 0.70)
```
        # dont_inherit: this module's `from __future__ import annotations` would turn the annotations into strings.
        exec(compile(source, "<provider>", "exec", dont_inherit=True), namespace)

        cache = TokenCache(namespace["provider"])
```
likely benign LLM, confidence 0.85 Test helper that execs a provider source string into a namespace to build a TokenCache.
Reasoning
exec of compiled source in test code, with a comment explaining the dont_inherit flag is needed to keep annotations from becoming strings. The source appears to be a test-generated provider definition, not an attacker payload. Non-literal exec in a test file is a legitimate reflection technique but worth a glance at where source comes from.
info ~~warn~~ → info by review dynamic-execution tests/lib/test_credentials.py:1902:14

`compile` called on a non-literal expression (ast pass, confidence 0.70)
```
        # dont_inherit: this module's `from __future__ import annotations` would turn the annotations into strings.
        exec(compile(source, "<provider>", "exec", dont_inherit=True), namespace)

        cache = TokenCache(namespace["provider"])
```
likely benign LLM, confidence 0.85 compile of a provider source string used by the exec in the same test helper.
Reasoning
compile call paired with the exec above; same test-helper context and explicit comment about annotation handling. No network, shell, or encoded payload involved.
info ~~warn~~ → info by review frame-introspection tests/lib/utils.py:51:13

`inspect.currentframe` — call-stack frame introspection (ast pass, confidence 0.70)
```
def get_caller_name(*, stacklevel: int = 1) -> str:
    frame = inspect.currentframe()
    assert frame is not None
```
likely benign LLM, confidence 0.90 Test utility retrieving the current stack frame to get a caller's name.
Reasoning
inspect.currentframe used to determine caller name for a stacklevel-based helper in test utilities. Standard introspection with a None assertion; no malicious use shown.
info ~~warn~~ → info by review dynamic-execution tests/test_utils/test_reflection.py:16:5

`exec` called on a non-literal expression (ast pass, confidence 0.70)
```
    # dont_inherit: this module's `from __future__ import annotations` would turn the annotations into strings.
    exec(compile(source, "<func>", "exec", dont_inherit=True), namespace)

    signature = signature_without_evaluating_annotations(namespace["func"])
```
likely benign LLM, confidence 0.85 Test helper that execs a function definition string to test signature reflection.
Reasoning
exec of compiled function source in a reflection test, with a comment explaining dont_inherit for annotations. The purpose is to test signature extraction without evaluating annotations; legitimate test scaffolding.
info ~~warn~~ → info by review dynamic-execution tests/test_utils/test_reflection.py:16:10

`compile` called on a non-literal expression (ast pass, confidence 0.70)
```
    # dont_inherit: this module's `from __future__ import annotations` would turn the annotations into strings.
    exec(compile(source, "<func>", "exec", dont_inherit=True), namespace)

    signature = signature_without_evaluating_annotations(namespace["func"])
```
likely benign LLM, confidence 0.85 compile of a function source string for the reflection test helper.
Reasoning
compile paired with the exec in the reflection test helper; same benign purpose and documented reasoning.

Reproduce This Scan

Run disclude locally to verify. Add --format sarif for SARIF output. Add --llm (with an API key in the environment) to include the LLM review.

cargo install disclude --version 4.1.0
git clone https://github.com/anthropics/anthropic-sdk-python anthropic-sdk-python && cd anthropic-sdk-python
git checkout v1.17.0
disclude scan --severity info --diff v1.16.0 .