~200K
living cells on the research chip
Research-stage · Berlin
The only thing that can stop a machine is a mind.
Synaspis is investigating whether living human neurons can learn to defend computer networks from attacks that change faster than software can adapt.
01
Biological network
Network defence
02
Closed-loop learning signal → action → feedback
Building toward first pilot experiments in 2027
See the study design01 / Premise
Cyberattacks now evolve at machine speed. Defence needs a new learning substrate.
Today's systems are powerful, but learning a new threat can demand vast amounts of data and repeated exposure. Biological neural networks learn differently: continuously, efficiently, and from direct interaction with their environment.
We are building the experiment to find out whether that difference matters for cybersecurity.
4
predefined experimental conditions
1
falsifiable question to answer first
02 / The first experiment
A network defended by neurons. Measured against AI.
We will connect a living neural culture to a simulated computer network. The culture receives information about the network, chooses defensive actions, and receives structured feedback. Then the attacker changes.
The question
Can a biological neural network adapt to a new attacker behaviour with less experience than a deep reinforcement-learning agent?
Network state
encode
Neural culture
decode
Defensive action
Fixed in advance
The outcome measure, controls, and analysis are defined before the first live run.
Fair comparison
Biological and artificial agents get the same network task and the same amount of experience.
Publish either way
Positive or negative, the result advances what we know about biological computing.
03 / Closed loop
How wetware defence works
A continuous loop translates network activity into a form the culture can sense, and neural activity back into actions the network can execute.
-
01
Show
Encode what is happening on the network as electrical stimulation.
-
02
Respond
Read the culture's activity through the electrodes beneath it.
-
03
Defend
Decode that activity into a move, such as isolating a compromised host.
-
04
Learn
Return structured feedback so the culture can adapt its future responses.
04 / Where we are
Built in simulation. Next: living neurons.
Complete
End-to-end software
The system runs on the Cortical Labs SDK simulator.
Complete
AI benchmark
The deep reinforcement-learning baseline is established.
Now
Study design & team
Finalising stimulation design, controls, and pre-registration.
Expected 2027
First live experiments
Run the study on CL1 cultures and publish the result.
05 / Questions, answered
The science is ambitious. The claims are not.
This is an experiment, not a finished security product. We will learn from the answer whether it supports the hypothesis or not.
Where do the cells come from?
The cultures are grown in a laboratory from induced pluripotent stem cells, which can be created from donated skin or blood cells. No brain tissue is involved.
Are the cultures conscious or able to feel?
The system uses a thin culture of roughly 200,000 cells, without a body, sensory organs, or brain architecture. There is no evidence that cultures of this scale are conscious or sentient. We still treat the ethical questions seriously and follow the standards of our research partners.
Does it work yet?
Not yet. The software works end to end in simulation; the hypothesis has not been tested on living neurons. The first experiment is designed to produce a clear answer.
Will living cells be connected to real customer networks?
No. The first study takes place in a controlled, simulated network. It tests learning behaviour, not production deployment.
Founder
Tomer Liran
Software engineer and two-time founder. Former network engineer in the Hellenic Navy. Building Synaspis in Berlin as part of Antler.
We are assembling the founding science team.
Build this with usWetware cybersecurity