Portugal Launches Open National AI Model, Europe Stakes Its Claim

AI & Machine Learning

Portugal has released its first open‑source national AI model in a government‑backed move aimed at boosting European AI sovereignty and supporting Portuguese language and public‑sector use cases. The project is pitched as an accessible alternative to US and Chinese commercial models, intended to give local researchers, companies and government agencies on‑ramps to build services without depending on foreign closed systems. By anchoring an open model inside Europe, Lisbon hopes to foster local innovation, improve data governance and make it easier for public institutions to adopt AI while remaining compliant with European rules. The release could also act as a test case for other European states considering national or regional open models as part of broader sovereignty strategies. Source: Reuters Verified: True

Anthropic has launched “Claude Science,” positioning the product as a research workbench that prioritizes integrated workflows — data, compute and reproducibility — rather than debuting a new foundation model. The offering bundles tooling and pipelines that make it easier to run experiments and manage compute with Claude models, addressing friction researchers face when assembling disparate infrastructure. Anthropic frames the move as targeting scientific and computational research teams that value end‑to‑end reproducibility and workflow ergonomics over raw model novelty. If it gains traction, Claude Science could shift some competition from model quality alone to the usefulness of developer and researcher tooling layered on top of models. Source: TechCrunch Verified: True

Consumer Hardware

No major stories this sector today.

Cybersecurity

Nissan disclosed that employee data was accessed following exploitation tied to the recent Oracle PeopleSoft zero‑day campaign, joining a growing list of organisations impacted by the fast‑moving chain. Nissan said it is investigating the intrusion with partners and assessing the scope of exposed employee information, underscoring how supply‑chain or widely used enterprise apps can amplify breach impact. The incident highlights the risk for companies that rely on legacy ERP systems and the need for rapid patching and threat hunting after public disclosures. Organisations using PeopleSoft are being urged to verify their exposure and apply mitigations immediately to prevent further data theft. Source: SecurityWeek Verified: True

A newly named “CitrixBleed” vulnerability affecting Citrix NetScaler appliances began being actively exploited almost immediately after public disclosure, with researchers reporting remote arbitrary memory reads from affected devices. The rapid weaponization demonstrates how internet‑exposed network infrastructure becomes a high‑value target and raises the urgency for organisations to patch or isolate vulnerable appliances. Researchers warned that unpatched instances could lead to data leakage and potential takeover, especially where NetScaler sits in front of corporate applications and VPNs. Administrators are advised to prioritize mitigations, apply vendor patches and audit external exposure to prevent further exploitation. Source: SecurityWeek Verified: True

Security teams observed a massive password‑spray campaign targeting the Azure CLI that generated tens of millions of login attempts, with reported volumes above 80 million attempts designed to discover weak credentials and abused service principals. The attack indicates automated, commodity adversary behavior shifted toward cloud developer tooling, amplifying risk to cloud accounts, CI/CD pipelines and service identities. Organisations are urged to enforce strong authentication controls, require MFA for service principals where possible, rotate credentials and monitor anomalous CLI activity and geographic login patterns. The scale of the campaign underscores that cloud‑native interfaces are now primary attack surfaces that defenders must instrument and protect. Source: SecurityWeek Verified: True

Google released security updates addressing 382 Chrome vulnerabilities, including multiple critical issues, in a coordinated remediation push that affects Chromium‑based browsers and downstream tooling. The unusually large batch reflects ongoing discovery efforts and coordinated disclosure processes, and Google urged immediate updates to reduce the window for exploitation. Browser vendors, enterprises and DevOps teams relying on embedded Chromium components will need to test and deploy patches quickly to avoid exposure, especially for critical CVEs. The bulletin is a timely reminder that keeping browsers and their runtimes up to date remains a frontline defense against diverse web and supply‑chain attacks. Source: SecurityWeek Verified: True

Enterprise Infrastructure

Reports indicate Meta is building a cloud business to sell excess AI compute and potentially models, aiming to monetise spare datacenter GPU capacity and compete with hyperscalers. Turning idle AI infrastructure into a commercial offering could reshape the enterprise AI compute market by adding a major new supplier with tight integration to Meta’s internal tooling and potentially attractive pricing for large workloads. The plan would let Meta recoup investment in GPUs and create new revenue streams, but it also raises questions about partner relationships with existing cloud providers and how customers will evaluate performance, SLAs and data governance. If executed, the move could intensify price and capacity competition for large model training and inference, forcing incumbents to respond on both pricing and specialised services. Source: Reuters Verified: True

Policy & Regulation

Cloudflare announced a new bot‑control policy that classifies crawlers into Search, Agent and Training categories and will by default block mixed‑purpose crawlers on ad pages, giving publishers until Sept. 15 to adopt the separation or face blocking/charging options. The change is intended to give publishers more control over scraping and a pathway to charge for content used to train AI systems, while constraining indiscriminate data collection by AI companies. Cloudflare’s policy creates a practical enforcement mechanism at the edge that could influence how large language model builders collect web data and negotiate with publishers for access. The August–September rollout window means publishers and AI firms must quickly update their crawler behaviours or risk being throttled, which could accelerate commercial arrangements for training data. Source: TechCrunch Verified: True

A UN advisory panel warned that AI developments are outpacing scientific understanding and government policy, saying unchecked progress could create catastrophic risks and calling for stronger international coordination on governance, safety evaluation and capacity building. The report stresses gaps in global oversight and argues for multilateral cooperation to align safety standards, testing and incident response across borders. Its conclusions add urgency to debates over export controls, shared testing facilities and investments in safety science, and could shape diplomatic efforts in upcoming forums. Governments and tech firms may face renewed pressure to accelerate responsible‑AI verification and cross‑border regulatory collaboration in response. Source: Reuters Verified: True