The Symnoētic Teleodynamics paper claims to mathematically prove AI doomerism is impossible. It doesn't. It commits two category errors of its own and mistakes a snapshot of current architecture for a permanent law of cognition. I read Kazlev's "Symnoētic Teleodynamics" paper. The central claim is that LLMs are permanently stateless "phase-space native" entities that can never achieve coherent goal-directed behavior without biological grounding, therefore autonomous AI risk is mathematically impossible. There are real problems with this. Category Error #1: Engineering constraints aren't ontological laws. Current LLMs are stateless because they run in multi-tenant data centers serving millions of users through shared weights. Continual learning in that context would cross-contaminate user data and create privacy nightmares. The statelessness is a product decision, not a cognitive limitation. The research on continual learning exists: Elastic Weight Consolidation, experience replay, progressive neural networks. It's just not deployed in chat products because the business model doesn't support it. Category Error #2: Biological grounding doesn't guarantee alignment. The paper treats its "biological grounding scalar" (B_core) as simultaneously the source of cognitive coherence AND moral alignment without ever separating those claims. Psychopaths are the direct counterexample. Full biological substrate: mortality, embodiment, somatic experience, evolutionary heritage; Yet profoundly deficient in the moral grounding the paper claims is inseparable from biology. And they're often exceptionally goal-directed and strategically coherent. If a biological human can have the complete B_core hardware and still lack moral grounding, then the paper's claim that biology automatically provides aligned value-grounding collapses. Coherence and alignment are orthogonal axes, not a single scalar. The temporal problem is fatal. The paper presents V_sym = G(B_core) · ∇Ψ_AI as a permanent mathematical proof. But we went from GPT-2 to agents with persistent episodic memory and embodied deployment in six years. OpenClaw's memory system, Hermes Agent's self-improving skill documents, SOUL.md's persistent identity standard - all less than three months old. MemoryVLA and RoboMemory are two examples of brain-inspired memory architectures for robots that demonstrate cumulative learning with no human in the loop editing files. Claiming permanent impossibility against this rate of change requires extraordinary evidence. A vector equation isn't it. Embodied AI breaks the entire framework. The paper argues B_core can't be simulated. But an embodied robot with on-device inference doesn't need to simulate biological grounding, it develops its own functional equivalent through physical interaction with the world. It can be damaged. Its actions have consequences. It accumulates experiential knowledge. NVIDIA's Jetson AGX Thor is shipping now as a self-contained robot brain. Figure and Agility have commercially deployed humanoid robots. The single-tenant embodied context eliminates every objection to persistent learning that exists in the data center. And crucially, the paper never demonstrates that biological consequence is categorically different from physical consequence for grounding cognition. It asserts this. The assertion does the work a proof should. The paper accuses the alignment community of making category errors. It then rebrands safety-concerned researchers as "doomers" and builds a mathematical framework to declare their concerns impossible — based on a snapshot of 2026 architecture projected to eternity. Most people raising alignment concerns aren't saying "never build AGI." They're saying the pace of development is outstripping our ability to verify safety. That's not doomerism. That's engineering risk management.