How modern EMFs may jam and miswrite the cell’s operating system
For most of the wireless era, the safety debate has been framed in one narrow way: if a signal does not ionize molecules and does not heat tissue beyond accepted limits, what serious biological effect could it have? That question is not wrong. At sufficiently high power, radiofrequency energy does heat tissue, and that thermal effect is real, measurable, and dangerous. But it is also an incomplete first question for a living system that is electrically patterned, redox-driven, and temporally organized across many scales.
The public-health evidence is also more complicated than either side usually admits. RF-EMF remains classified by IARC as “possibly carcinogenic to humans” (Group 2B), yet a 2024 WHO-commissioned systematic review found moderate-certainty evidence that mobile phone use likely does not increase the risk of several of the most studied adult head tumors. At the same time, the U.S. National Toxicology Program reported clear evidence of malignant heart schwannomas and some evidence of malignant gliomas in male rats under high-exposure conditions, while the Ramazzini Institute also reported increased heart schwannomas in male rats under long-term far-field GSM-like exposure. Meanwhile, a 2024 systematic review of oxidative-stress biomarkers judged the evidence for or against RF-EMF effects to be overall very low certainty and highly inconsistent. That is not a settled all-clear, and it is not a settled catastrophe either. It is a sign that the thermal-only story may be incomplete, while the non-thermal story is still under mechanistic construction.
My proposal starts from a different picture of the cell. A cell is not just a bag of molecules obeying local chemistry. It is a stateful system. It has electrical gradients, membrane potentials, redox variables, calcium dynamics, structural memory, and genome topology. In Michael Levin’s bioelectricity language, cells and tissues use distributed voltage patterns to store and process morphogenetic information. In the ceLLM language I have been developing, those same systems can be understood as a form of probabilistic state integration: fast inputs are interpreted through slower structural priors. That does not mean cells run software large language models. It means biology appears to solve inference problems using wet, multiscale hardware.
That distinction matters because it changes how we think about DNA. In the old blueprint metaphor, DNA is passive text. In the ceLLM metaphor, DNA topology is closer to a physical prior. The sequence still matters, but so do compaction, accessibility, geometry, and local molecular context. For mitochondrial DNA in particular, TFAM compacts a roughly 5 μm contour-length genome into nucleoids on the order of 100 nm, and recent work shows nucleoid accessibility and packaging are heterogeneous rather than fixed. In plain English: mtDNA is not just code. It is a dynamic physical object whose state may shape how mitochondria interpret stress, energy demand, and signaling inputs. Those are the cell’s “weights and biases” only in a metaphorical sense: evolution has already trained the system physically, not digitally.
The healthy control plane: light, redox, voltage, and structure
Once you stop treating mitochondria as mere ATP factories, a different architecture comes into view. Mitochondria sit at the intersection of energy production, reactive oxygen species, calcium handling, membrane potential, retrograde signaling, and gene-expression change. They are not peripheral batteries. They are a control layer. Reviews on mitochondrial signaling increasingly frame them as hubs that link oxidative stress, inflammation, aging, and cell fate.
Part of that control layer is optical. Ultra-weak photon emission, or UPE, is not fringe. It is a documented byproduct of oxidative metabolism. Living cells emit a few to several hundred photons per second per square centimeter, typically in the ~200–800 nm range, and the dominant chemistry is ROS-linked formation of excited singlet oxygen and carbonyl species. In eukaryotic cells, mitochondria appear to be central sources of that emission. That does not prove UPE is a language. But it firmly establishes that mitochondria both generate and sit inside a field of extremely weak, state-dependent optical activity.
External photobiology already gives us the first half of the story. In mammals, circadian photoentrainment depends on melanopsin-containing intrinsically photosensitive retinal ganglion cells, which relay environmental light information to the brain’s clock circuitry. In photobiomodulation research, cytochrome c oxidase is still widely treated as a principal mitochondrial photoacceptor or central response node for red and near-infrared light. So the mainstream field is already comfortable with the idea that light can regulate physiology through specialized retinal sensors on one side and mitochondrial chromophores on the other.
What your framework adds is the missing internal half: the possibility that mitochondria are not only receivers of external light but also samplers of their own endogenous optical-redox weather. That idea gets more interesting because it now has empirical footholds. In 2023, Mould and colleagues reported non-chemical signaling between chemically isolated mitochondria. In 2024, Kalampouka and colleagues showed that 734 nm near-infrared exposure increased senescence selectively in the cancer cell lines they tested, with parallel changes in mitochondrial ROS and membrane potential. That is not yet proof of an internal photonic network, but it is exactly the kind of bidirectional “read/write” behavior a mitochondrial control-plane hypothesis would predict: mitochondria emit state-dependent signals, and external light can shift mitochondrial state.
Inside the cell, the most plausible architecture is layered, not singular. The source layer is ROS chemistry and mitochondrial chromophores. The receiver stack includes ETC chromophores, crista membrane architecture, calcium-handling machinery, and perhaps the mtDNA nucleoid itself. The distribution bus may involve cytoskeletal proteins: electronic energy migration over nanometer scales has been measured in microtubules, and actin waves are now known to help maintain mitochondrial content mixing and organelle homeostasis. Biomolecular condensates then provide a candidate memory layer by modulating local electrochemical equilibria and redox reactions. Put differently: the cell may be less like a one-part antenna and more like a nested photonic-redox motherboard.
That is where mtDNA becomes interesting. The point is not that mtDNA is literally a little radio loop in free space. Hydrated organelles are lossy, screened, thermally noisy environments. The point is that mtDNA is a topology-sensitive physical object embedded precisely where redox chemistry, calcium flux, membrane potential, and ultra-weak light all intersect. Recent THz work makes that framing harder to ignore: a 2026 ACS Nano study reported that 34.5 THz, but not 36.1 THz, enhanced mitochondrial biogenesis through calcium influx and the PGC-1α–NRF1/2–TFAM pathway. That result does not prove mtDNA is the receiver. It does show that frequency-selective perturbations in the right band can map onto mitochondrial outcomes and TFAM-linked transcriptional readouts.
Where modern EMFs may enter: the dual threat
This is where I think your “dual threat” framing is strongest. The first threat is jamming. If biology uses weak optical, redox, and electrochemical signals to estimate state, then fidelity matters. The second threat is miswriting. If external fields perturb control nodes upstream of calcium, membrane potential, or mitochondrial redox state, then a cell can update itself based on distorted inputs. The key insight is that the biological consequence would not need to look like a burn. It would look like degraded timing, degraded coordination, degraded repair, and degraded tissue-level coherence. That is a systems claim, not a claim that every router is a carcinogen.
The most cautious way to name the mechanistic hypothesis is S4-Mito-Spin. “S4” refers to membrane voltage-sensor machinery. The S4 helix is indeed the canonical voltage sensor in many voltage-gated ion channels. “Mito” refers to the downstream mitochondrial layer, where calcium, ROS, membrane potential, and transcriptional adaptation are tightly coupled. “Spin” refers to radical-pair-sensitive photochemistry, especially cryptochrome/flavin chemistry. But the evidentiary status of those layers is not equal. The existence of S4 voltage sensing is textbook channel biology. The centrality of mitochondria to calcium-redox integration is mainstream. Radical-pair chemistry in cryptochrome is real, but its relevance differs sharply by organism and context.
That nuance matters, because one of the weak points in online EMF discourse is overstatement. A voltage-gated-calcium-channel route for RF effects has been proposed, but ARPANSA’s review concluded that currents induced at guideline limits are many orders of magnitude too low to affect gating directly and that experimental studies have not validated RF effects on calcium transport into or out of cells. So the honest statement is not “we already know Wi-Fi forces S4 open.” The honest statement is: an S4-first route is a candidate mechanism, but it is not established at ambient guideline-level exposures. If such a route exists, it likely depends on specific boundary conditions the generic exposure literature has not yet nailed down.
The cryptochrome layer deserves the same discipline. In Drosophila, the circadian clock shows light-dependent magnetic sensitivity through cryptochrome, consistent with radical-pair physics. In mammals, cryptochromes are unquestionably core clock proteins, but whether they act as physiologically relevant magnetoreceptors remains unresolved. The most defensible way to use cryptochrome in this framework is as a co-Zeitgeber hypothesis: not the main mammalian clock input, which is retinal and melanopsin-led, but a possible secondary route by which abnormal magnetic environments could degrade timing fidelity in certain tissues or molecular milieus. That is interesting. It is not yet settled mammalian chronobiology.
Where the framework becomes more biologically plausible is through density gating. If external perturbations act by biasing fragile control nodes rather than by uniformly “poisoning the whole body,” then vulnerability should be local. Tissues dense in ion-channel machinery, mitochondria, flavin/heme chemistry, or timing-sensitive bioelectric coordination may be more susceptible than inert tissue. That does not prove a mechanism for the NTP or Ramazzini tumor patterns. But it does provide a principled reason not to expect uniform effects across all tissues and all signals. Frequency, modulation, cell state, time of day, and tissue architecture may all matter.
Why senescence, cancer, and morphology belong in the same conversation
The recent senescence literature is a strong test case because it sits at the boundary between metabolism and fate. Senescent cells are not dead; they are state-shifted. Kalampouka’s line of work is intriguing precisely because it places biophoton emission, near-infrared responsiveness, ROS, calcium, and membrane potential inside one phenotype. Your framework’s value is not that it “proves” why those results happen. Its value is that it organizes them into a mechanistic hierarchy: optical/redox source, topology-sensitive receiver stack, calcium amplification, and transcriptional memory. If that hierarchy is wrong, the experiments can break it. If it is right, senescence becomes a readable and writable output of mitochondrial control-plane fidelity.
The regenerative-bioelectricity literature points the same way from a different angle. Levin and colleagues have argued for years that bioelectric networks carry instructive information about growth and form. Emmons-Bell et al. showed that transient gap-junction blockade in Girardia dorotocephala could induce temporary species-specific head morphologies before the worms later remodeled back toward their native anatomy. That is exactly the kind of result that makes a multi-timescale control-plane concept worth taking seriously: fast electrochemical perturbations can create stable but reversible anatomical states. Your addition is to ask whether mitochondrial photonic-redox fidelity is one of the hidden substrates supporting those larger bioelectric envelopes.
What policy looks like if cellular fidelity matters
The policy conclusion is not that society must choose between the internet and biology. It is that we should stop designing connectivity as if RF were the only wireless layer worth scaling. Even without accepting the strongest version of your hypothesis, a precautionary systems strategy is already justified by uncertainty, congestion, security needs, and architectural flexibility. That means fiber-first backhaul, wired defaults for fixed high-duty devices, adaptive low-power and low-duty-cycle modes where possible, nighttime network hygiene, and exposure transparency that goes beyond a single SAR number. Exposure is architecture.
This is where Li-Fi becomes practical rather than rhetorical. IEEE 802.11bb-2023 now provides an interoperable Li-Fi standard with bidirectional operation from 10 Mb/s to 9.6 Gb/s. Real deployments and trials are no longer hypothetical; NHS England’s published case work on Li-Fi and IoT describes benefits in reliability, speed, security, low latency, and avoidance of RF congestion in device-dense environments. Li-Fi is not a total replacement for RF. It is line-of-sight constrained, room-bounded, and best thought of as a complementary local-area layer. But that is exactly why it matters: it gives us a path toward network pluralism instead of RF monoculture.
So the right political ask is not “ban Wi-Fi.” It is: build Li-Fi compatibility into the next generation of local networks, luminaires, and IoT standards; make fiber and wired backhaul the default whenever practical; treat radio as a resource to be used deliberately, not sprayed indiscriminately; and fund mechanistic studies that can actually adjudicate between thermal, membrane, chromophore, and topology-based models. That is a sane engineering agenda whether the final answer turns out to be modest or profound.
What RF Safe can contribute
RF Safe is strongest, in my view, when it is presented not as a slogan but as a systems-engineering discipline. The project matters when it forces attention onto the variables that real biology and real network design both care about: source power, duty cycle, distance, body position, nighttime exposure, modulation, local device density, backhaul choice, room architecture, and user control. Its most defensible public message is not “we have already solved the mechanism.” It is “the organism is more electrically and optically organized than the thermal-only debate assumes, so network design should respect biological fidelity instead of pretending the only meaningful variable is bulk heating.”
And that is where ceLLM becomes more than a metaphor. In ceLLM terms, the cell integrates fast variables such as redox bursts, photon statistics, calcium spikes, and membrane shifts through slower priors such as nucleoid compaction, contact-site density, chromophore abundance, condensate state, and transcriptional memory. Put in simpler language: the cell is always estimating what world it is in. A high-fidelity environment lets those estimates converge. A low-fidelity environment does not need to “blast” the system to hurt it. It only needs to bias it often enough, or at the wrong nodes, to erode timing, coordination, and repair.
What would make this more than a compelling story
The good news is that this framework is testable. If TFAM or topoisomerase perturbation never shifts optical or narrow-band THz sensitivity, the mtDNA-centered version weakens. If rho0 depletion leaves structured optical responses intact, simpler chromophore-first explanations gain ground. If calcium blockade erases everything, the model narrows toward membrane/amplification layers. And if none of the proposed topology-sensitive predictions survive careful replication, then the right thing to do is to say so. A strong theory is not one that resists falsification; it is one that tells you exactly how it could fail.
That is why I think this line of thought is worth publishing and worth reading. It is not asking the public to believe in magical light. It is asking biology to take its own multiscale organization seriously: photons, redox chemistry, calcium, voltage, topology, and memory all inhabit the same cell. The question is not whether those layers exist. They do. The question is whether we have been underestimating how tightly they are coupled — and whether modern network environments can degrade that coupling in ways our current safety language barely knows how to describe. If your framework is right, the next era of public health will not be built only around chemistry and heat. It will be built around fidelity.
