Introducing GPT-Red
An internal automated red teamer on a mission to find our models’ prompt injection vulnerabilities at scale, helping us build stronger defenses before wider deployment.
openai.com/index/unlocking-s…
CyberSeQ retweeted
The real question in agentic AI is not whether the agent can do the work. It is where human accountability has to stay in the loop. We are building PostQ as an end-to-end agentic cryptographer. One product, one workflow, covering the full lifecycle: discovery, roadmapping, migration, validation, and red-team / pen-test. We are not anti-agent. We are pro-accountability. Cryptography is one of the few enterprise domains where every decision carries audit weight, regulatory weight, and breach-liability weight, and that is exactly why the design question is not "how much can the agent do" but "where does the human expert have to remain the signer." A few things that are true about the current AI wave, and that we take seriously. A handful of startups have genuinely compressed development cycles by 5 to 10x with serious investment in AI and agents. Small teams now ship in months what used to take a year. Agents code, test, debug and automate continuously, with no shift boundary. Token cost flexes in a way that headcount does not, and it scales faster than hiring. A few things that are equally true, and that the noise often skips over. Agents do not carry accountability. When they fail, root-causing the failure is hard, sometimes intractable. In sensitive domains, robotics, security, system architecture, customer trust, cryptography, a human still has to own the call. Token cost is not a fixed line item either. It can spike, and depending exclusively on a small number of model providers is a real concentration risk. So the real question is not "human or token." It is sharper than that. Which parts of the workflow should be fully automated, and which parts need a human who can be held responsible for the outcome? That is the question we want to answer with PostQ, in the open, alongside the people who do this work for a living. If you are a cryptography consultant with a live client engagement, PQC migration, CBOM/QBOM, key inventory, FIPS readiness, anything in that surface area, and you are open to testing PostQ on a real project under NDA, I would like to hear from you. The goal is not to prove the agent can do everything. It is to map, together, where the agent ends and the expert begins. Better tools for the people who already carry the responsibility. That is the bet.
CyberSeQ retweeted
Once the first RSA/ECC instance is broken by CRQC, the scaling becomes unpredictable. This isn’t the Bombe era, it’s around 2029, and we’re facing a convergence of advanced technologies. This is my last awareness tweet, from here I focus on how to do it.
CyberSeQ retweeted
Hot-off-the-press: Researchers outline a new method of running machine learning on quantum comptuers, and open-up the opportunity for its application to cybersecurity applications and botnet detection. Just published in Nature Scientific Reports: nature.com/articles/s41598-0…
CyberSeQ retweeted
Quantum computers are here and now, and they allow for some applications to be run in a massively pararel way. So here's some new thinking on the use of quantum machine learning and botnet detection ... link.springer.com/article/10…
CyberSeQ retweeted
Standing on the Horizon of One of the Greatest Advancements in Science: The Quantum Computer
Meet Hybrid Quantum Machine Learning (HQML) for Cybersecurity medium.com/asecuritysite-whe… Internationally leading work from @madjid_tehrani
CyberSeQ retweeted
Here is some of our latest work ... Standing on the Horizon of One of the Greatest Advancements in Science: The Quantum Computer - Meet Hybrid Quantum Machine Learning (HQML)
Standing on the Horizon of One of the Greatest Advancements in Science: The Quantum Computer - Meet Hybrid Quantum Machine Learning (HQML): medium.com/asecuritysite-whe…












