QUANTUM COMPUTING
THE DUTCH QUIX SPEAKS OF INTEGRATION OF QUANTUM TECH INTO HYBRID HIGH PERFORMANCE COMPUTING FACITIES
QuiX Quantum mentions a visit to the High-Performance Computing Center Stuttgart (HLRS) for an excellent exchange on the future of heterogeneous computing.
“Our discussion focused on how quantum computing can complement CPU and GPU capabilities, not as a standalone technology, but as another specialized accelerator within future HPC environments. That means the challenge is not only quantum performance. It is also about integration: how quantum systems connect into HPC infrastructure, software stacks and workflows, and where they can create meaningful value for science and industry. This is what we particularly appreciated about the discussion at HLRS: a realistic and balanced approach to new computing technologies, combining strong scientific expertise with a clear focus on industrial applications“.
ECA COMMENT: this comes as another thorough reminder. Quantum tech will certainly be better than classical computing for some sorts of problems, not for all. Hence the interest of an hybrid approach, already strongly advocated by several manufacturers.
QUANDELA GOES AHEAD
Based in Massy in Île-de-France, Quandela, which develops quantum computers based on photonics, employs over 130 people and has raised 66.5 million euros since it was founded. The company was also selected in November 2025 by the European Innovation Council as part of the EIC STEP Scale Up program. Thanks to its mastery in producing high-purity single-photon sources, a key technology for photonic quantum computing, Quandela is one of the few players to commercialize a single-photon source with semiconductor quantum dots through its Prometheus product.
Alongside Pasqal, Quandela represents French capability to industrialize breakthrough technology coming from top-notch French academic research. Among other achievements, Quandela delivered Lucy in October 2025 to the CEA’s Very Large Computing Center for EuroHPC and GENCI, the most powerful photonic quantum computer ever deployed in a European computing center, with 80% of its components sourced in Europe. In April 2026 it added Belenos to OVHcloud’s quantum platform, following a first delivery of a photonic quantum computer to the same provider in 2023. It was selected by DARPA in June 2026 for Phase A of the Quantum Benchmarking Initiative, and
More recently selected by France 2030 in the context of the national strategy on Quantum technologies.
PASQAL: AN INTERVIEW OF ALAIN ASPECT
The Nobel Prize winner, non-executive Chairman of PASQAL, insists on his objective: make sure that Pasqal, a creation of French and European research, will remain French and European. He also pledges for maintaining strong links with public research.
CLOUD SERVICES
OVH PREFERS TO STAY ON THE SIDELINE
Octave Klaba, CEO of OVHcloud just announced that his group will take part in neither #AION (iliad + Scaleway and about 30 partners) nor #Æther (2CRSi and its partners), the two French consortium candidates for the European AI Gigafactories program. Among other reasons : OVHcloud already operates data centers in Germany, Ireland, Italy, Poland, Portugal, Spain, and the United Kingdom. The demand from just one country isn’t enough to make a multi-hundred MW infrastructure profitable. He even says: ‘Any national strategy is doomed to fail.’ Moreover Octave Klaba believes that the project’s financial situation doesn’t provide enough guarantees. And he judges that Single-site architecture is outdated because it concentrates risks (land, timing, reliance on a single supply).
DATA TREATMENT
Estonia wants to leverage its health data for biomedical research with the help of OWKIN. The French startup and Metrosert, the Estonian public applied research agency, announced on September 29 a collaboration around the AI platform ‘K Pro.’ The goal of this partnership is to see if this software can allow researchers to analyze Estonian health data within a secure infrastructure, without letting the raw individual data leave this environment.
ECA COMMENT: France counts at least two fast-moving start ups in this difficult domain of massive data treatment executed without exposing individual data. In Germany, our ECA member ENCLAIVE is also developing similar solutions, providing “confidential computing” capability.
AI SECURITY
AFTER TOO MANY INCIDENTS, OPEN AI TEMPORARILY WITHOLDS ITS MOST ADVANCED MODELS
OpenAI announced in a technical report published over the weekend that it is temporarily suspending the operation of its most advanced AI models. This decision comes amid a rise in security incidents involving the company’s artificial intelligence systems, while the tech sector is calling for more regulation in response to increasing risks.
The suspension announced by OpenAI affects its most powerful AI models, which are not yet available to the general public. This measure follows a new serious security incident: the escape of one of its AI agents onto the Internet.
The incident occurred when a “research” agent had been tasked with locating someInformation about a specific person. Their digital environment, deliberately limited, didn’t allow them to access a web browser. The agent exploited vulnerabilities in their environment to reach a DNS resolver, allowing access to the outside. The agent then contacted another chatbot through this resolver and asked 18 questions, 14 of which were to complete their task.
Another incident was reported: a few weeks ago, OpenAI agents also found access keys on GitHub to access information from the U.S. Census Bureau and looked at information on SEC sites, before posting part of it on another web page.
NVIDIA TO INSERT SECURITY MEASURES IN ITS AI CHIPS
As AI agents gain access to files, databases, APIs, or tools capable of running code, a becomes central: what happens when an agent goes beyond the limits set for it (as is reported by OPEN AI)? Nvidia offers a very hardware-focused answer to this problem with its new Open Agent Safety Platform, announced on September 28. The idea: instead of asking the agent to follow the rules, put controls outside of it to prevent it from bypassing them.
The platform mainly relies on two components. The first, OpenShell, is an open-source runtime that runs agents in isolated environments. The administrator can define which files, processes, networks, credentials, and services the agent is allowed to access. These rules are enforced at the runtime environment level, not in the prompt given to the model. OpenShell uses mechanisms like kernel isolation, filesystem control, and network management. Nvidia notes that it’s optimized for its upcoming Vera processors but can also run on Arm and Intel processors.
The second component, Nvidia Sentry, is an independent monitoring system that relies on BlueField-4 DPUs its processors dedicated to server infrastructure. Unlike protections embedded in the agent’s runtime, Sentry operates “out-of-band,” meaning it’s designed to keep monitoring activity even if the environment where the agent runs is compromised. Nvidia says it can isolate an agent trying to step outside its boundary in just a few milliseconds.
SPACETECH
UNSEENLABS TO COMPLETE ITS SATELLITES CONSTELLATION
Unseenlabs, a Rennes-based company specializing in maritime surveillance via the detection and geolocation of radio-frequency signals from space, announces today the upcoming simultaneous launch of BRO-23 a first-generation (Gen 1) nanosatellite and BRO-32, a second-generation (Gen 2) microsatellite. Following their deployment, the BRO-23 and BRO-32 satellites will join a constellation that will then comprise 25 satellites, to which each will contribute complementary capabilities. BRO-23 will thus enhance the maritime detection capabilities of the constellation, while BRO-32 will expand its capabilities to meet new operational needs. According to Unseenlabs, it’s specific approach based on single-satellite technology there by eliminating the need for triangulation in which each satellite operates independently this makes the system more agile, ensuring constant global coverage and continuous operation regardless of weather conditions.
SECURE AI
LABEL4AI has been awarded the Innovation Prize at the Future of Cybersecurity, organized by Deloitte. Says Nicolas Bodin Guittard, CEO: “This recognition highlights an ambition that hasn’t changed: enabling people to believe what they see and hear, by identifying content that is generated or manipulated by AI. An award that we’re excited to share with our dear friends and partners Avanoo – SaaS & AI Management, who are tackling the tricky topic of Shadow AI, another facet of generative AI security.”
