The obvious concern would be data-dependent backdoors for malicious “decoding”, i.e. correctly decoding ordinary data, but manipulating the decoding of targeted data in some compromising way.
That relies on some rather far-fetched assumptions about what the attacker might reasonably be able to control undetected, and what goals they might reasonably be able to achieve through such low-level data corruption.
Maybe information leakage? Tweak some low-order float bits in the decoded results with high-order bits from data the decoder recognizes as “interesting”?
What's the attack vector in this case? The Wasm is loaded from the file itself. If they can compromise the file then its cheaper to just compromise the data directly.
What I’m imagining is essentially a supply chain attack: The victim (mistakenly) trusts the attacker to supply an encoder. The encoder appears to function normally, but in fact will subtly leak information smuggled in decoded values of the victim’s data.
That relies on some rather far-fetched assumptions about what the attacker might reasonably be able to control undetected, and what goals they might reasonably be able to achieve through such low-level data corruption.
Maybe information leakage? Tweak some low-order float bits in the decoded results with high-order bits from data the decoder recognizes as “interesting”?