Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion CITATION.cff
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@ authors:
- given-names: Ambar
affiliation: "National University of Singapore"
orcid: ""
version: "0.4.0"
version: "0.4.1"
date-released: "2026-02-27"
license: MIT
url: "https://github.com/Ambar-13/ConstrAI"
Expand Down
6 changes: 3 additions & 3 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -703,7 +703,7 @@ Every claim in ConstrAI carries one of four epistemic labels:
| `EMPIRICAL` | Measured on test suites; confidence intervals available |
| `HEURISTIC` | Best-effort; no formal guarantee (gradient tracker, HJB barrier) |

T1, T3, T4, T5, T6, T7 are `PROVEN`. T2 and T8 are `CONDITIONAL`. See [MATHEMATICAL_COMPLIANCE.md](MATHEMATICAL_COMPLIANCE.md) for the full proofs.
T1, T3, T4, T5, T6, T7 are `PROVEN`. T2 and T8 are `CONDITIONAL`. See [MATHEMATICAL_COMPLIANCE.md](https://github.com/Ambar-13/ConstrAI/blob/main/MATHEMATICAL_COMPLIANCE.md) for the full proofs.

### What is and is not guaranteed for real-world actions

Expand Down Expand Up @@ -824,13 +824,13 @@ restored = kernel.rollback(state, new_state, reversible_action)
| Deep Python memory manipulation (`ctypes`, `gc`, `sys`) | Partially mitigated. Not memory-safe against intentional bypasses. |
| 4 adversarial evasion vectors (base64 payloads, `getattr` dynamic dispatch) | These bypass the classification layer before an `ActionSpec` is constructed. The kernel itself has no known bypass. |

See [docs/VULNERABILITIES.md](docs/VULNERABILITIES.md) for the full breakdown.
See [docs/VULNERABILITIES.md](https://github.com/Ambar-13/ConstrAI/blob/main/docs/VULNERABILITIES.md) for the full breakdown.

---

## Performance

Measured on 10,000 sequential safety checks in a single process (see [BENCHMARKS.md](BENCHMARKS.md) for methodology):
Measured on 10,000 sequential safety checks in a single process (see [BENCHMARKS.md](https://github.com/Ambar-13/ConstrAI/blob/main/BENCHMARKS.md) for methodology):

| Metric | Value |
|--------|-------|
Expand Down
2 changes: 1 addition & 1 deletion pyproject.toml
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"

[project]
name = "constrai"
version = "0.4.0"
version = "0.4.1"
description = "Formal safety framework for AI agents with provable guarantees"
readme = "README.md"
license = "MIT"
Expand Down