MORE ON ANTHROPIC’s DECISION TO POSTPONE THE COMMERCIAL LAUNCH OF CLAUDE MYTHOS – WHAT IT SAYS OF PRECAUTIONS TO TAKE WHEN IMPLEMENTING AN AI SYSTEM
As our readers know, ANTHROPIC has decided to restrict the use of CLAUDE MYTHOS to some selected partners, mainly in cybersecurity, for a while. beyond the issues on cybersecurity MYTHOS has uncovered a lot of zero day vulnerabilities as well as some very old but yet unknown, hence triggering a fear that hackers could exploit it faster than defenders this decision should help stakeholders wishing to deploy an AI system take a breath and set up sanitary measures. First, the user must clearly decide what the system will be up for the user, not the vendor or the IT integrator and on this occasion think of criteria to follow up the results : are they in line with the expectations ? Second, what about the user’s security systems ? Can they face a sudden discovery of undetected defaults ? Third, is the user able to detect and mitigate “autonomous unwished behavior” of the AI system? This refers to the fact that, during its first runs, MYTHOS automatically tried to erase proofs of modifications it had organized in the IT system ! (Nota: these measures come in echo to a post by Jean-Philippe Clair, from ASI)
More generally, the decision by ANTHROPICto so far restrain CLAUDE MYTHOS distribution sounds as a welcome reflection. Keith King, a former White House scientific adviser, comments by suggesting it’s more than time to think on the consequences before unconsciously deploying AI systems. At least he pledges for a much stricter governance by all stakeholders. “The situation also illustrates how AI development is entering a phase where capability is outpacing established safeguards. As models become more powerful, their ability to interact with and analyze complex systems introduces new categories of risk that extend beyond traditional misuse scenarios. This raises questions about how to balance innovation with responsibility in an environment where the stakes are increasingly systemic”.
This statement comes as Thomson Reuters Foundation says that, according to its surveys, 87% of organizations adopting AI do it without any specific governance.
SPACETECH
SATELIoT, a Spanish satellite communication operator, has secured a 100 M€ additional funding. The company, which has already put 6 new gen satellites on Low Orbit, plans to launch 16 more ones. It delivers 5G New Radio services of IoT, voice, and video communication.
SOVEREIGN COLLABORATIVE SUITES
Two more new collaborative suites, to contest the supremacy of US vendors: EURO-OFFICE, developed and supported by a coalition between IONOS (a proud member of the ECA), Nextcloud, Eurostack, XWiki, OpenProject, Soverin, Abilian et bTactic and a second one launched by SOPRA STERIA, which can run either on a SecNumCloud or on prem.
CLOUD: EU COMMISSION AT LAST LEANS TOWARDS SOVEREIGNTY
We hereby publish a PR by the EU Commission: “The European Commission is strengthening the European Union’s digital sovereignty by awarding a tender that allows EU institutions, bodies, offices and agencies (Union entities) to procure sovereign cloud services for up to €180 million over 6 years.
The four awarded providers are European companies: Post Telecom Luxemburg with its partners CleverCloud and OVHcloud [these free providers are joined in the DEEP Consortium], StackIT, Scaleway, and finally Proximus, which partners with S3NS (a joint venture of Thales and Google Cloud), Clarence and Mistral.
This tender supports the Commission’s broader efforts to enhance its own sovereignty, reinforcing strategic control across key technologies and infrastructure.
The awarded providers were selected based on their alignment with the Commission’s Cloud Sovereignty Framework, which measures sovereignty across eight objectives. These include strategic, legal, operational, and environmental considerations, as well as supply chain transparency, technological openness, security, and compliance with EU laws. The Commission awarded four contracts in parallel to ensure diversification and resilience, avoiding over-reliance on a single provider. To be eligible, the providers had to reach rigorous assurance levels ensuring that non-EU third parties have limited control over the technologies the providers use, or services they provide.”
QUANTUM COMPUTING: OVH CLOUD TO EMBARK A QUANDELA PLATFORM
QUANDELA’s BELENOS 12 qubits platform will be proposed in “as a service” mode for research, algorithms testing, learning systems, and also for real use cases.
