The Ledger
Paper models runtime checks against smart-contract type confusion
A new paper formalizes type confusion, in which hostile code supplies values that do not behave as their advertised types. Its authors prove security results for runtime checks in a theoretical model that includes smart contracts and other decentralized systems.

A formal model for hostile inputs
A formal model of type confusion links several classes of attacks—including reentrancy and confused-deputy attacks—to mismatches between the types trusted code expects and the behavior of values supplied by an adversary. The preprint, “Type-Confusion Security,” by Silei Ren, Hanxi Chen, Suraaj Kanniwadi and Andrew C. Myers, sets out a security property and runtime checks for an information-flow-typed system.
The arXiv record lists it as a v1 submission on Nov. 15, 2026. Its focus is not a new smart-contract product or a test of deployed contracts, but a formal analysis of how trusted code can interact with attackers who ignore its type rules.
Checks at the trust boundary
The authors define type-confusion security as a comparison between two settings: one where attackers must obey the type system and another where they can supply ill-typed values. Their stated security goal is that the unrestricted attacker should gain no more power than the attacker constrained to well-typed code.
The paper describes an information-flow-typed core calculus for decentralized systems that communicate through remote procedure calls. Within that model, the authors prove that runtime checks can preserve the security guarantees made by the type system even when attackers supply ill-typed inputs. They mechanize the proof in Lean.
One example involves a token exchange service. In the paper’s scenario, a user supplies a payment callback for the service to call; an attacker instead supplies the exchange’s own payment function. That substitution could lead the exchange to pay the user rather than collect payment. The paper describes a runtime check at the function call: the trusted callee checks whether the call comes from a context its security rules permit.
The authors also discuss reentrancy, where a callback re-enters trusted code while a function is waiting on an untrusted call. Their model extends an authority-lock mechanism to enforce reentrancy security, and the paper proves security properties for both confused-deputy and reentrancy attacks within its formal setting.
Broader applications, formal scope
Smart contracts are one example in the paper, alongside cross-chain protocols, federated systems and Model Context Protocol ecosystems. Its appendix describes an MCP host that invokes external tools: a substituted callback could use a privileged tool executor without separate approval, while a reentrant callback could attempt to use an authorization more than once.
The work’s results concern its defined calculus and proof framework.
Sources
- Original source · Preprint Type-Confusion Security arXiv:2611.20298
Related stories
- Lean Proof Caps EIP-7002’s Block Call at 210,000 Gas
- XRPL Foundation outlines first phase of formal verification
- Researchers Mutate Smart-Contract Tasks to Catch Copied Answers
- Curve says fee-burner flaw let attacker take about $38,200
- Paper Finds ZK Rewards Raised Attackers’ Share in Federated Learning Tests