Loading vulnerability…
CVE-2025-67644 is an SQL injection vulnerability in LangGraph SQLite Checkpoint (langgraph-checkpoint-sqlite) affecting versions 3.0.0 and earlier. The flaw is in the checkpoint search/listing logic, specifically the _metadata_predicate() path used when querying checkpoints by metadata. User-controlled metadata filter keys are interpolated directly into SQL via f-strings, including within json_extract expressions, without validation or safe parameterization. Because the injection point is the metadata filter key rather than only the value, an attacker who can supply untrusted filter keys can break out of the intended JSON path context and alter the SQL statement. Reported exploitation includes arbitrary SQL manipulation and UNION-based injection to return attacker-controlled fake checkpoint rows from the SQLite database.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
Repository contains a single Python proof-of-concept exploit script plus a README. - Purpose: Demonstrate CVE-2025-67644, a SQL injection in langgraph-checkpoint-sqlite (<3.0.1) where attacker-controlled *metadata filter keys* are unsafely interpolated into a SQLite JSON extraction expression used by SqliteSaver.list()/alist(). - Main file: `LangGraph SQLite Checkpoint SQL Injection.py` (entry point via `main()` with argparse). It opens a SQLite checkpoint DB using `SqliteSaver.from_conn_string(db_path)`, optionally seeds demo checkpoints via `create_dummy_data()`, then: 1) Lists all checkpoints with no filter. 2) Runs a legitimate filter `{ "user_id": "alice" }`. 3) Triggers SQLi using `malicious_filter = {"env') OR '1'='1": "dummy"}` to bypass filtering and return all rows. 4) Optional output controls: `--dump-all` prints checkpoint + metadata; `--threads-only` prints only thread_ids. - Capabilities: authorization/filter bypass and data enumeration/exfiltration from the checkpoint database (thread IDs, checkpoint content, metadata). No RCE or network scanning; it is a local/application-level PoC that relies on a vulnerable application path that passes untrusted filter keys into the library. - Structure: 2 files total (1 code file in Python, 1 README). README documents affected versions, vulnerable code pattern, example injection key, and remediation (upgrade to >=3.0.1 / key regex validation).
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
11 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
A SQLite injection vulnerability in LangGraph's checkpointer component that can be part of a vulnerability chain leading to remote code execution.
A SQL injection vulnerability in LangGraph's SQLite checkpoint implementation that can be used as part of a chain to achieve remote code execution.
A SQL injection vulnerability in LangGraph's SQLite checkpointer where attacker-controlled filter dictionary keys are interpolated into a JSON path expression, enabling SQL injection and, when chained with unsafe deserialization, remote code execution.
A SQL injection vulnerability in LangGraph's SQLite checkpointer that can be triggered via user-controlled filter keys in get_state_history()/list() metadata queries. It is significant because it can be chained with unsafe msgpack deserialization to achieve remote code execution.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.