Human biocomputing: security and organoid intelligence in the new AI era

By Kike Gandia · AI & Security · 21 min read

Organoid intelligence (OI) is emerging as a new frontier in biological computing, connecting lab-grown mini-brains to digital systems to create “biocomputers” that can learn and process information. The approach taps into the efficiency and plasticity of living human neural networks: the human brain can perform on the order of an exaflop (10^18 operations) on just ~20 W, while the most powerful supercomputer (Frontier) needs ~21 MW for comparable performance (a million times more power). Given the energy and scalability limits of conventional computing, OI promises more powerful and more sustainable computation. Below we look at what organoid intelligence is and how far it has come, the pioneering companies and their business models, practical applications, cybersecurity implications, technical considerations, and future scenarios, with recommendations for engaging with the field in an informed way

~20 W: used by the human brain for on the order of an exaflop (10^18 operations)
~21 MW: needed by the Frontier supercomputer for comparable performance
1,000,000×: more power drawn by silicon than by living tissue for the same work

The energy argument for growing neurons to compute.

What is organoid intelligence (OI) and what progress has been made?

Brain organoids grown on a chip with microelectrodes for stimulation and monitoring. Organoid intelligence (OI) is an emerging field that seeks to “harness the innate biological capabilities of brain organoids for computing and synthetic intelligence by interfacing them with digital technology.” In essence, brain organoids (tiny spheres of human neurons derived from stem cells) are grown on microelectrode arrays and connected to equipment that can electrically stimulate the neurons and read their signals in real time. Thanks to advances in bioengineering and microfluidics, these 3D organoids mimic basic functional aspects of the human brain (e.g., synapse formation and spontaneous electrical activity), laying the groundwork for “brains-on-a-chip” for biological computing.

In recent years we have seen the first glimmers of organoid learning. One milestone was DishBrain: in 2022, a culture of 800,000 human and mouse neurons in a dish learned to play the video game Pong, the first time brain cells in a dish had carried out a goal-directed task. The experiment, published in Neuron, showed that neurons could be shaped through stimulation to produce adaptive behavior (“responding with something like intelligence,” according to the lead author). It also opened the door to testing how external factors affect biological intelligence: the team planned, for example, to observe how alcohol impairs DishBrain's Pong performance, as an analogy for intoxication in a living organism. These advances suggest that organoids can learn and store information, albeit only rudimentarily for now.

In fact, the computing capacity of today's organoids is modest. FinalSpark researchers reported in 2023 that, through repeated stimulation, their platform had managed to reliably store 1 bit of information in an organoid — a “Bbit,” or biological bit, by analogy with the first qubit in quantum computing. A single bit may sound trivial, but it is a crucial proof of concept: it shows that a living neural network can be programmed (trained) to remember a state. From there, the plan is to scale up complexity by increasing the number of neurons and the lifespan of the cultures, and by using advanced machine learning algorithms to optimize electrical stimulation. Experts predict it will take at least one or two decades for OI to reach the complexity of a small animal's brain. Still, there is already a scientific roadmap: in 2022 the first international OI workshop brought universities and companies together to sketch out a 10-year plan to establish OI as a discipline, with a view to transforming computing, neurological research and drug development

2022 — DishBrain learns to play Pong
800,000 human and mouse neurons in a dish carry out a goal-directed task for the first time.
2023 — The first biological bit
FinalSpark reliably stores 1 bit of information in an organoid through repeated stimulation: the “Bbit,” the counterpart of the first qubit.
1-2 decades — The complexity of a small animal's brain
How long experts predict it will take organoid intelligence to reach that level, by scaling up neurons, culture lifespan and stimulation algorithms.

From a video game to a single bit: the milestones that mark where the field really stands.

Watch the video on YouTube

Key OI companies: FinalSpark, Cortical Labs and their platforms

The rapid progress of biocomputing has been driven by specialist startups that combine neuroscience and technology. Two stand out: FinalSpark (Switzerland) and Cortical Labs (Australia/Singapore), pioneers with complementary approaches:

Switzerland FinalSpark: Grows human organoids on chips and sells remote access to these “biological processors” by subscription: biocomputing as a service, via a Python API.
Australia Cortical Labs: Builds CL1 devices: living neurons on chips with built-in life support and an interface for connecting sensors and running learning algorithms.
United States Koniku: Neural biosensors that pair olfactory neurons with chips to detect explosives or pathogens: “olfactory cyborgs” that mimic a dog's nose.

Three approaches the pioneers are taking to the same problem.

FinalSpark has developed the Neuroplatform, billed as the first living computing platform available in the cloud. FinalSpark grows human organoids on chips and offers 24/7 remote access to these “biological processors” over the internet. Users (academic or corporate researchers) can connect via a Python API to organoids in FinalSpark's lab, sending stimuli and reading neural responses in real time, all from a digital notebook interface (Jupyter, for example). This biocomputing-as-a-service model is sold by subscription: in early 2024 it cost $500 per user per month for shared access to four organoids, with discounts or free access for certain academic projects. FinalSpark reported strong early demand — 34 universities asked to use its biocomputers — although its current infrastructure supports only a handful of simultaneous users (it has run around 7 groups in parallel, and has tested more than 1,000 organoids over 4 years of continuous operation). FinalSpark's business model centers on providing a turnkey R&D environment: the company handles cultivation, maintenance (automated microfluidics, temperature and gas control, etc.) and data storage (it has generated 18 terabytes of electrophysiological data), while the customer focuses on the experiments. In short, FinalSpark sells organic compute time in the cloud, much like a GPU cloud but with “brains.”
Cortical Labs, for its part, is turning OI into a dedicated hardware product. Its flagship is the CL1, described as “the first code-deployable biological computer.” The CL1 is a shoebox-sized device that combines a silicon chip with living cultured neurons on top, kept in a controlled environment (nutrient solution, temperature) so they can survive for up to 6 months. This hybrid hardware–wetware system (hard silicon + organic tissue) includes every component needed: a built-in life-support system, sensors and electrodes for stimulation and readout, local data storage and a touchscreen for monitoring the state of the “neural network” in real time. Developers can run code directly on the integrated neurons, thanks to Cortical's biological operating system (biOS), which creates a simulated environment for them. For example, information from a virtual environment (such as a game) can be sent to the neurons and their spikes read back to see how they react and learn. The CL1 runs in closed loop, meaning the neurons interact bidirectionally with the software in real time, which enables reinforcement learning experiments (the neurons' actions feed back into the simulated environment). Cortical Labs offers the CL1 both for direct sale to labs (with an SDK and support) and through Cortical Cloud, a cloud service that lets anyone without the physical device upload code remotely to real neurons hosted on Cortical's own CL1 units. In that sense, Cortical Cloud is similar in concept to FinalSpark's Neuroplatform, providing on-demand access to neural computing without any in-house infrastructure. Cortical Labs has raised investment (~$11M) to bring the technology to market, focusing on sectors such as pharmacology (animal-free in vitro testing), neurological research and, eventually, bio-inspired AI solutions. Its Pong demonstration with DishBrain gave it global visibility, and with the CL1 it now aims to scale those capabilities commercially.

Cortical Labs CL1 devices: each unit contains living neurons grown on chips, with built-in life support and an interface for connecting sensors and running learning algorithms. Alongside FinalSpark and Cortical Labs, there are other notable players in biocomputing. The startup Koniku (US), for instance, specializes in neural biosensors: it combines neurons (derived from olfactory cells) with electronic chips to detect chemicals such as explosives or pathogens, creating “olfactory cyborgs” that mimic a dog's nose. Although its focus is different (more biosensing than general-purpose computing), it shares the same fusion of biology and electronics. Several academic institutions (Johns Hopkins, Monash, etc.) and public consortia are also driving OI research, often in collaboration with these startups. Taken together, an emerging OI ecosystem is taking shape: FinalSpark offering OI-as-a-Service, Cortical Labs selling ready-to-use OI hardware, and others exploring specific niches (biological sensing, neuromorphic interfaces, etc.). Their business models range from monthly subscriptions and equipment sales to intellectual property in training algorithms and potential deals with industries (e.g., pharmaceuticals) interested in applying OI.

Watch the video on YouTube

Current and potential practical applications

Neurological and medical research: Brain organoids provide in vitro human models for studying the brain without resorting to animals or patients. Teams at Johns Hopkins, for example, use organoids to compare the neural networks of autistic and neurotypical people, in order to understand the neural basis of the condition and test drugs in a controlled environment. Similarly, DishBrain's creators plan to use the system to study the effect of drugs (such as alcohol) on cognitive function. This suggests that OI could speed up drug discovery and toxicology testing in neurology, offering an ethical alternative that is more relevant than traditional animal models. Its use in neurodegenerative diseases (Alzheimer's, epilepsy) is also being explored, by creating patient-derived “mini-brains” on which to trial therapies.
Artificial intelligence and advanced computing: In the long run, bioprocessors could complement or outperform silicon chips at certain AI tasks. Neurons have an innate capacity for generalization, unsupervised learning and adaptation that today's algorithms can barely emulate. A well-trained OI system could, in theory, recognize complex patterns using little energy and little training data, which would make for more efficient machine learning. Potential areas that have been mentioned include sensorimotor learning (robots controlled by organic networks), unstructured data analysis, language translation and other areas where the human brain excels. Although there are no established commercial AI applications yet, companies such as Cortical Labs envision use cases in sectors such as customer service (systems that “learn” in a more human way) or cybersecurity (for example, threat detection through adaptive neural patterns).
Modeling complex systems: Another futuristic promise is to apply organoids to modeling complex phenomena such as the climate, financial markets or other systems with many variables. The reasoning is that biological neural networks can, in principle, process information in parallel and handle uncertainty more flexibly than deterministic algorithms. If organoids with millions of neurons can be achieved, they could serve as biological co-processors for simulations where classical computing is inefficient. This vision, while speculative, is part of the appeal of OI: in the coming decades biocomputing is expected to complement traditional computing across multiple domains, much as quantum computing targets specific very-high-performance niches.

Right now, OI applications are mostly experimental and niche. Even so, they are already proving their worth: DishBrain showed how neurons can act as intelligent agents in a game; FinalSpark already serves labs that analyze neural data for research; Koniku has prototype biological sensors in airports. Each small achievement validates a use case. As the technology matures (for example, organoids that live longer than 6 months, with higher cell density and refined training protocols), we are likely to see the first concrete commercial applications. These could include drug-screening platforms using personalized organoids for pharmaceutical companies; security systems that can “smell” hazardous substances; or bio-inspired AI modules built into data centers to cut the energy consumed by massive training runs. The possibilities are broad, but realizing them will depend on solving technical challenges and showing that OI can deliver tangible advantages over existing solutions.

Risks and opportunities from a cybersecurity perspective

As with any disruptive technology, organoid intelligence brings cybersecurity risks and opportunities that CISOs and security architects need to consider:

Potential risks

A wider attack surface: An OI system adds a biological layer to the infrastructure, introducing unconventional attack vectors. A malicious actor could, for example, try to sabotage a biocomputer by tampering with its environmental conditions (temperature, nutrients) through IoT controls, causing the organoid to malfunction or die. Or they could send malicious electrical stimulation patterns to corrupt the neural network's learning (analogous to feeding adversarial inputs to a conventional AI). Traditional protection mechanisms (firewalls, authentication, etc.) will have to adapt to account for the unique physical and logical variables of living systems.
Data integrity and reliability: In critical systems, entrusting decisions to a biological “black box” makes validation difficult. If an organoid computes a result (e.g., detecting an anomaly), how do you verify that it has not been compromised or biased? The stochastic nature of neural activity makes reproducible audits hard. This calls for continuous monitoring of the organoid's health and state (for instance, detecting abnormal drift in its signal patterns) to make sure it is still operating within expected parameters and has not been tampered with.
Espionage and industrial sabotage threats: Organizations that develop or use OI could become targets of cyberattacks aimed at stealing intellectual property (e.g., effective stimulation algorithms, biomedical research data) or sabotaging projects. APT campaigns against bioengineering companies are not far-fetched; one recent report anticipates that advanced actors could target companies using organoid bioreactors or 3D tissue printers, seeking to damage the equipment or quietly alter the cultures. Cyberbiosecurity is emerging as a discipline to protect this intersection of digital and biological systems, with measures such as access control for automated labs, network isolation for biomedical equipment and encryption of biometric data.
Ethical and regulatory implications: While this article takes a technical and corporate angle, it is worth noting that using human tissue for computing raises ethical questions that could lead to regulation. From a security standpoint, new legal frameworks could require additional audits and controls (e.g., ensuring traceability of the cells used, informed donor consent and prevention of unauthorized use). CISOs will need to work with compliance to ensure that their organization's OI systems are not only technically secure but also compliant with future digital bioethics rules (thereby avoiding penalties or reputational damage).

Security opportunities and advantages

Bio-inspired defenses: Paradoxically, OI could provide novel tools for defensive security. Neurons are experts at picking out signals from noise and adapting to changing patterns; in a cyber defense role, a trained organoid could detect subtle intrusions or zero-days in network traffic that slip past systems built on rigid rules. Imagine an IDS (intrusion detection system) powered by a living neural network that “sniffs out” anomalous behavior in data packets much as our brain detects dangerous stimuli. Although the idea is still theoretical, companies such as Cortical Labs suggest that integrating “biological intelligence” with conventional computing could break new ground in areas such as adaptive security protocols.
Resilience to algorithmic attacks: Today's attackers exploit predictable vulnerabilities in conventional software and hardware (buffer overflows, adversarial attacks on ML models, etc.). An OI system, by its chaotic and non-deterministic nature, could be less predictable and therefore harder to “hack” with known scripts. Crafting an adversarial input that fools a convolutional neural network is feasible, for example, because we know the gradients of its artificial neurons; fooling a real organoid could be substantially harder, since its exact connections are unknown and change over time. This does not mean OI is inherently secure, but it does add diversity to the technology stack, which in security is desirable to avoid single points of failure. A heterogeneous ecosystem in which some critical workloads run on biocomputers could, in the long run, improve overall resilience against cyberattacks, much as biodiversity slows the spread of pests.
New authentication and encryption methods: On the offensive and defensive security front, some researchers speculate about using biomolecular properties to strengthen cryptography — for example, generating encryption keys from an organoid's unique activity patterns (impossible to replicate without the same tissue) or using living tissue as an “oracle” for hash functions. These uses are still at the edge of science fiction, but they illustrate the intriguing security questions raised by bio-digital convergence. Either way, security leaders would do well to keep a close eye on cyberbiosecurity, which is already exploring both how to protect biological data and how to harness biological systems for digital protection.

In short, the arrival of organoid intelligence forces companies to extend their threat model into domains previously reserved for biotech labs. CISOs will need to plan for both logical controls (security of the APIs that control organoids, segmentation of the networks where devices like the CL1 run, identity management for scientists with remote access) and physical ones (access to wet-lab facilities, environmental sensors). At the same time, it makes sense to examine how these technologies could feed into the defense strategy, while keeping a level head: neither underestimating the risks because the technology is new nor giving in to unfounded alarmism. As with AI and quantum computing, the key will be to bring security into OI pilot projects early, so that hardening, monitoring and incident response guidelines suited to this new class of systems can be developed.

Technical considerations

From a technology architecture perspective, integrating OI platforms involves specific technical challenges that CTOs and architects need to assess:

Scalability and performance: Unlike conventional servers, biological “nodes” (organoids) do not follow Moore's Law. A typical organoid today has somewhere between 50,000 and a million neurons, equivalent to the neural network of an insect or a small fish, and its activity is thousands of times slower than a GHz CPU. Scaling performance means increasing the number of organoids and/or the number of electrodes per organoid, which biology limits (beyond a certain size, the tissue can become necrotic). FinalSpark, for example, has replaced more than 1,000 organoids in 4 years to keep its platform running continuously — each organoid wears out after weeks or months of use. This reality imposes a different scaling paradigm: instead of doubling transistors, tasks will have to be parallelized across multiple organoids and longer training times accepted. CTOs should plan on the basis that, at least in the short term, OI is a scarce resource that does not scale linearly (you can't just “add more RAM” — growing new tissue takes time and specialist staff). OI companies are researching biotechnologies to extend organoid lifespan (>100 days is typical) and increase cell density, but this will require continued scientific progress.
Integration and connectivity: The good news is that current platforms have been designed with integration into common development environments in mind. Both Neuroplatform and Cortical Cloud offer high-level APIs (a Python SDK) and browser access, so scientists and engineers can work with organoids using familiar notebooks and data science tools. It is possible, for example, to call methods that send electrical pulses to an organoid and receive data streams from its neurons, all from a Python script. This plug-and-play approach extends to the hardware: the CL1 exposes standard USB ports for connecting external sensors (cameras, actuators) and can be integrated into smart lab environments. For an architect, this means a biocomputer could be treated conceptually as just another component in the hybrid cloud: a service that receives requests (input) and returns results (output) via REST APIs or an SDK, with the caveat that it needs more generous latencies and protocols suited to streaming neural data. It is crucial, however, to make sure connectivity is properly segmented. A cloud-exposed organoid must operate behind strong security barriers, and any physical OI device connected to the corporate network should be isolated in VLANs or OT networks, since it runs specialized software (e.g., Cortical's biOS) that may be more vulnerability-prone at this early stage.
Maintenance and operations (DevBioOps): Adopting OI means bringing wet-lab processes into IT environments. Maintenance of a biological server, for example, includes replacing culture media, calibrating microfluidic pumps, disposing of spent organoids in a biosafe way, and so on. FinalSpark gets around this by hiding the complexity behind its service (the company runs the organoids 24/7, monitoring temperature, pH and nutrient flow in a closed loop). An organization that opts for its own hardware (such as the CL1) will have to train staff or contract support for these BioOps tasks. New operational questions arise: how do you version and roll back the “state” of a trained organoid? (There is no git for synapses, but its activity can be recorded so that a replacement organoid can eventually be retrained.) How do you manage biological variability? Two organoids will never be identical, so applications must tolerate some divergence in results and may need to retune parameters when switching biological chips. CTOs will need to work with scientists to develop MLOps best practices adapted to OI (continuous training, periodic calibration, A/B testing across organoids, etc.), perhaps coining the term DevBioOps for the seamless integration of software development cycles with the management of biological systems.
Human-machine interfaces and visualization: Working with miniature brains generates massive volumes of real-time data (multidimensional electrical waveforms). Specialized observability tools will be needed so that IT and science teams can understand what is happening inside the organoid. Adapted SCADA-style dashboards are already in use: FinalSpark, for example, logs its organoids' data in a time-series database (InfluxDB) and provides live visualization of metrics such as neural firing, applied currents and nutrient flow. Architects should plan to integrate these feeds with their existing monitoring platforms (for instance, sending alerts to SIEM systems if an organoid shows abnormal inactivity that could indicate a failure). Operators will also need to learn new performance indicators: beyond CPU, RAM and disk, what matters here is the vitality of the tissue (e.g., baseline spike rate, cellular stress indicators). In short, observability infrastructure will have to expand to include biological telemetry, possibly relying on AI to interpret complex patterns in neural data.

Overall, adopting organoid intelligence will require close collaboration between IT, security and data or life scientists. From an architecture standpoint, the recommendation is to start with controlled pilot projects in sandbox environments to build in-house expertise before scaling. The learning curve is steep — bringing “living matter” into computing poses both technical and cultural challenges — but the companies that get there early will gain an edge in know-how and intellectual property in this emerging space.

Future scenarios and recommendations

The new AI era may well have human biocomputing at its core, as a complement to silicon and quantum technologies. Looking ahead, we can sketch out a few scenarios and offer recommendations for business leaders:

The technology matures (5-10 years): In the optimistic scenario, the next decade will bring gradual but solid progress in OI. Organoids could scale to tens of millions of neurons, approaching the complexity of a small mammal's brain, thanks to improvements in biofabrication (e.g., 3D printing of neural tissue) and in hybrid training techniques (using machine learning to guide optimal stimulation). Interface standards for biocomputers are likely to emerge, perhaps driven by academic-industrial consortia like the one formed at the 2022 OI workshop, defining common stimulation protocols, neural data formats and even biocomputing performance benchmarks. For companies, this will mean a safer ecosystem for adoption — with several compatible vendors, documented best practices and use cases proven in production. Recommendation: stay current by taking part in OI communities (workshops, consortia), encourage internal R&D teams to run proofs of concept with organoids (for example, by partnering with universities or startups on pilot projects), and build up interdisciplinary in-house talent (bioinformaticians, engineers with a grounding in neuroscience) who can judge soundly when and how to integrate OI into corporate solutions.
Adoption in high-value niches (5-15 years): Some industries will probably adopt OI sooner than others. The pharmaceutical/biotech and defense sectors are already showing interest: the former to speed up drug discovery and in vitro toxicity testing, the latter for a competitive edge in new forms of intelligence and sensing. Adoption could also come in quantitative finance (seeking any computational edge in complex modeling) or telecommunications (green data centers with lower power consumption). In national security, it is conceivable that agencies will experiment with OI for signals intelligence analysis. Recommendation: investors and executives in the relevant sectors should track startups and scientific advances that could upend their fields. The global biocomputing market, valued at ~$9.4 billion in 2024, is projected to exceed $100 billion by 2037 (CAGR ~19.9%), pointing to rapid growth and potential strategic investment opportunities. Assessing possible technology partnerships or acquisitions now can put your organization ahead of the curve. At the same time, prepare for eventual disruption: just as AI forced a rethink of business models, OI could shift paradigms in the medium term.
Convergence with AI and traditional computing: Biocomputers are unlikely to replace digital computers entirely; rather, they will coexist in hybrid architectures. A data pipeline could, for example, be preprocessed on conventional GPUs, hand off certain cognitive tasks to an OI module and then return the results to a classical system for integration. This interconnection will require specialized standards and middleware. Big tech companies could eventually enter the field and offer “organic AI” services on their cloud platforms once the science is sufficiently proven. Recommendation: CTOs should start picturing future system designs in which components of very different kinds (CPUs, neuromorphic ASICs, organoids, etc.) coexist. On the security side, start drafting policies for assessing OI vendors (security audits of their labs, certifications) and require unconventional SLAs (how do you translate 99.9% availability into an environment where the “machine” is a living thing?). Gradually build provisions for these emerging technologies into IT strategy plans and corporate research roadmaps.
Ethical challenges and social acceptance: Further out, if OI produces systems with advanced cognitive abilities, debates will arise about consciousness, rights and the ethics of their use. From a corporate standpoint it is relevant to anticipate the reputational risk. Just as the ethical use of AI is scrutinized today, society may in future question the use of “mini-brains” for commercial purposes. Recommendation: be proactive about involving ethics committees and ensuring transparency in any OI initiative. Communicate clearly the measures taken (for example, using cells not linked to identities, approval by bioethics committees, compliance with regulations). Be ready to answer questions from the media or regulators about how you ensure these technologies are used responsibly. Public trust could become a key factor in the viability of biocomputing-based businesses.

In conclusion, human biocomputing with organoid intelligence is an exciting frontier that combines the efficiency of the brain with the precision of machines. For CISOs, CTOs and investors, today's challenge is to separate hype from reality: the achievements are still nascent (one bit stored, one game learned), but the trajectory points to profound effects on how we compute and solve problems. Experts are unanimous in recommending early engagement — getting informed, experimenting on a small scale and contributing to the conversations that will shape standards and policy — so as not to fall behind in what could be the next computing revolution. Just as the shift from vacuum tubes to semiconductors marked a turning point, putting living neurons into our systems may seem futuristic today, but forward-looking organizations are already incubating it in their roadmaps.

Sources: FinalSpark, Cortical Labs, Johns Hopkins University, Monash University, Frontiers in Science, Pace Ventures (Medium), among others. (See the full article for detailed references.)

Related service

AI and LLM penetration testing

Keep reading

Audit our AI systems