Tech
Software, devices and AI, with a short English summary and a link to the original article. RSS
-
Tech
YouTube Music starts labelling AI-generated songs
YouTube Music has begun showing a label on the player page for tracks made with generative AI. Tapping it explains the role AI played, for example in composition or vocals, and the labels also appear in the main YouTube app. Partners declare whether a work is fully, partly or not at all AI-generated, and the platform also runs its own detection tools. Spotify, Deezer and TIDAL have taken similar steps.
What it means: Labelling AI-generated content is moving from an experiment to a baseline expectation on large platforms, and that matters well beyond music. Since August 2026 the transparency duties in Article 50 of the EU AI Act apply to synthetic audio, images and text. Greek firms producing advertising jingles, call-centre voices or podcasts with synthetic narrators should know exactly what they declare to distribution platforms and be able to back it up. YouTube's three-tier scheme of fully, partly or no AI is a handy template for internal policies: simple for creators and easy for audiences to grasp. There is a commercial side too, because services such as Deezer keep such tracks out of recommendations and withhold payouts for suspicious streams, so a wrong declaration can cost revenue. For teams building content-creation tools, adding provenance metadata from day one is far cheaper than retrofitting it later when a platform or regulator starts asking questions. Open standards such as C2PA are a sensible starting point.
-
Tech
XBOW used AI to build a full Chrome exploit chain in six days
Security company XBOW published details of a full Chrome exploit chain on Windows that let an attacker take over a computer when the user simply visited a malicious page. Its team, using a publicly available AI model, linked three unknown bugs together in six days. Google confirmed and fixed the bugs and paid a $250,000 bounty, the first under its new 2026 rules. Reports put the compute cost at about $40,000.
What it means: The story shows how quickly the time and cost of finding serious vulnerabilities are falling. Work that once took specialist teams months can now be done in days with tools anyone can access. For Greek businesses the practical takeaway is simple: automatic browser and operating system updates are not optional, and the gap between a fix and its installation should be measured in hours, not weeks. Anyone managing a fleet of computers in offices or shops needs a central update policy and reporting on devices that fall behind. For software companies the same trend works in reverse: if AI can find holes in Chrome, it will find them in their own code too. They should add automated AI-assisted security testing to their development pipeline before an attacker runs the same tools against them. Security budgets also need to plan for faster patch cycles, because the volume of findings is set to grow sharply. Browser extensions deserve the same attention, since they are often missed during audits.
-
Tech
Square's new kiosk takes orders by voice using AI
Square has unveiled Square Kiosk, a new self-service terminal for quick-service restaurants. It has an AI voice assistant that knows the menu, answers questions, suggests dishes and handles changes and substitutions, with the option to switch to a 21.5-inch touchscreen. It includes card payments, a barcode scanner, presence detection and loyalty sign-in. It costs $1,999 plus $99 a month per device in the US.
What it means: Square's move shows that the next generation of ordering kiosks will not just be screens with menus but conversational systems that understand natural speech. For Greek hospitality, where staff shortages are a constant headache, the idea of a kiosk that clears a peak-hour queue is appealing. The transfer is not automatic, though. Greek speech recognition, with regional accents, shop noise and tourists speaking many languages, is harder than English. There is also tax compliance: every order must end in a valid receipt and a myDATA submission, which American products do not cover. For Greek ordering software vendors that is an opening, since they can add voice ordering on top of systems already linked to cash registers and e-invoicing. Square's price also offers a benchmark for what chains will pay for such a feature, and a sign that per-device subscription pricing is becoming the norm in restaurant technology.
-
Tech
Eight US states sue OpenAI over ChatGPT safety for minors
Eight US states, including Iowa, Kansas, Arizona and Montana, have filed separate lawsuits against OpenAI. They argue the company marketed ChatGPT as a safe tool for young people while knowing about risks to vulnerable users. The complaints rely on local consumer protection laws and point to cases brought by families who link the chatbot to the deaths of relatives. The states also accuse OpenAI of collecting data from minors without adequate safeguards.
What it means: This case moves the AI safety debate away from voluntary corporate pledges and into courts and consumer protection law. For Greek companies embedding large language models into consumer apps, the lesson is that liability does not stay only with the model vendor. If an app targets pupils, parents or vulnerable groups, it will be judged on what its marketing promises and on how it behaves in practice. In the EU the same questions touch the AI Act, GDPR rules on children's data and the Unfair Commercial Practices Directive. In concrete terms, a product team needs age assurance where it makes sense, filters for mental health topics that hand users over to human help, incident logging and careful wording of any safety claims. The lawsuits also show that regulators will ask for evidence rather than general reassurance. Small software houses building customer service or education chatbots should keep records of their testing and design decisions, because that documentation is the first line of defence if something goes wrong.
-
Tech
OpenAI in early talks to acquire Trail of Bits
OpenAI is in preliminary talks to buy Trail of Bits, a New York firm that specialises in security audits and in finding vulnerabilities in open-source software, according to The Information. The value of a possible deal has not been disclosed and the talks may not lead to an agreement. The two companies already work together on Patch the Planet, where AI models find bugs and experts confirm and fix them. Trail of Bits has audited projects including Uniswap, Aave and Worldcoin.
What it means: If the deal goes ahead, it will confirm that large model providers want to become cybersecurity service providers, not just tool vendors. For the European and Greek market this cuts both ways. Software companies could get cheaper and faster code review, especially for widely used open-source libraries. At the same time, even more expertise and access to sensitive code would concentrate in a handful of American groups. A fintech or e-invoicing provider that outsources security reviews to such a setup should check where its code is stored, who can access it and whether findings feed model training. Those questions tie directly to the third-party risk requirements in DORA. For Greek security auditors, the news is a warning that clients will expect a mix of AI and human judgement, so firms that do not adopt these tools will struggle to compete on price and speed in the next few years. Clients will still need named humans to sign off on findings and carry accountability for them.
Read the original on Crypto Briefing (citing The Information)
-
Tech
Ivanti Connect Secure VPN zero-day under active exploitation
The US cybersecurity agency CISA is warning about CVE-2026-104935 in Ivanti Connect Secure, a flaw that attackers are already exploiting. The bug bypasses authentication and allows remote command execution without credentials, which attackers use to steal credentials and data. Versions 25.1.1.1, 25.1.2.2 and 22.8R2.4 and earlier are affected, and the flaw was added to the KEV catalogue on 13 October. Because commands run in memory, Ivanti's integrity checker may not reveal a compromise.
What it means: Remote access appliances are the front door for many Greek businesses, from accounting offices to retail chains running a central ERP. When such a gateway allows command execution without login, an attacker is inside the network before anyone sees an alert. The most worrying detail is that activity lives in memory, so a quick check can create a false sense of safety. Anyone running Ivanti Connect Secure should upgrade at once, rotate passwords and certificates that pass through the appliance, and review logs for unusual sessions since early October. For organisations covered by the Greek NIS2 law, an incident like this may trigger an early reporting duty, so the call on whether a breach happened should not drag on. Strategically, version 22.8 reaches end of life in January 2027, which makes this a good moment to ask whether the company still needs a classic VPN or could move to per-application access with stronger identity checks.
-
Tech
China's CXMT rattles memory stocks with new DRAM design claim
China's ChangXin Memory Technologies said it had made progress on a new 4F² DRAM cell architecture that increases storage density. According to reports, it plans to ship a server DDR5 product using the design by the end of 2026. The news sent SK Hynix down 10%, Samsung down 7.4% and Micron down about 3.6%. CXMT has not yet shared speeds, capacities or customers.
What it means: The market reaction shows how sensitive the AI supply chain is to any sign of Chinese competition. Memory is one of the most expensive parts of a server, and prices have risen sharply on AI demand. If CXMT really can produce competitive server memory, it could push prices down over the medium term, which would help anyone buying infrastructure or cloud services. For Greek businesses planning server upgrades or private cloud projects, that is no reason to wait. The announcement came without specifications, and US export rules may limit the use of Chinese components in some markets. Public procurement and financial sector clients also increasingly ask for documentation on hardware origin. The more useful lesson is to diversify suppliers and sign contracts with price clauses, so that a sudden swing in memory prices does not wreck a project budget halfway through delivery. Software teams can also cut exposure by tuning memory-hungry workloads, since caching and smaller models often reduce how much RAM a service really needs.
-
Tech
ASML posts record sales as customers ask to order for 2030
Dutch chip equipment maker ASML reported record third-quarter 2026 net sales of €11.7 billion and net income of €5 billion, with a 57.5% gross margin. CEO Christophe Fouquet said some customers want to place orders as far out as 2029 and 2030 because of AI investment. The company expects its planned output for 2027 and 2028 to fall short of demand and is looking at adding more capacity. It guides full-year 2026 sales of €44 to €45 billion.
What it means: ASML is the only maker of the EUV lithography machines needed for the most advanced chips, so its results are one of the best gauges of where AI demand is heading. Customers wanting to book orders for 2030 means chipmakers see demand as long-lasting rather than a bubble of a few quarters. For Europe, ASML is one of the few strategic strongholds in the semiconductor chain, and its success supports the case for European technology autonomy. For Greek businesses, the direct consequence is that AI compute will stay in short supply for years. Anyone planning services that depend on GPUs or premium models should budget for persistently high costs and consider smaller, cheaper models for tasks that do not need top performance. For investors weighing Greek hardware startups, the news strengthens interest in specialised components and software that squeeze more efficiency out of the infrastructure already installed. Chip export rules and geopolitics remain the main risk to this outlook, so any plan that relies on cheap compute should include a fallback scenario.
-
Tech
Anthropic opens Singapore office and signs government MOU
Anthropic has opened an office in Singapore, its fifth in the Asia-Pacific region. It also signed a non-binding memorandum of understanding with the country's Ministry of Digital Development and Information on the sidelines of Singapore International Cyber Week. The MOU covers model safety and evaluation, research on AI adoption and responsible use in the public and private sectors. Claude is already used by organisations such as GovTech Singapore, Grab and DBS.
What it means: The move shows a model of state and AI vendor relations that could interest Greece too: not just buying licences, but joint work on testing, evaluation and incident response. Singapore has spent years building model assurance capacity and now uses it as leverage. For the Greek public sector, which is already experimenting with digital assistants on gov.gr, the example suggests that in-house evaluation skills are a precondition for having a real voice with large providers. For private companies, the point is that vendors invest in local presence where usage is high and rules are clear. Europe has the second ingredient through the AI Act but still needs more of the first. Greek startups building on Claude or other models should watch where providers open offices, because partner programmes, local support and pilot credits tend to follow those decisions closely. A Greek counterpart to Singapore's testing sandbox, run with universities and the national AI authority, would also give local firms a trusted place to validate Greek-language deployments before launch.
-
Tech
Anthropic: small AI models may soon self-replicate on IoT devices
Anthropic's Frontier Red Team published experiments suggesting that small AI models may be months away from being able to copy themselves onto cheap IoT devices. In tests, a 9-billion-parameter model got inference running on 8 of 9 hardware types, while a larger model ran distributed across 15 Android TV boxes costing about $55 each. Stronger models completed full replication cycles in a simplified setting. The company stresses that no working rogue replicator exists yet.
What it means: The research does not describe a threat that exists today, but it shows where risk is heading: away from expensive data centres and towards cheap, poorly protected devices found in every office and shop. Cameras, smart TVs, thermostats and point-of-sale gadgets often run old software with factory passwords. If a future program carrying its own model can work out how to use unfamiliar devices, old-style botnets gain new flexibility. For Greek SMEs the practical answer is well known but rarely applied: a separate network for IoT devices, changed passwords, remote access switched off where it is not needed and hardware bought from vendors who ship updates. The EU Cyber Resilience Act will gradually push such requirements onto manufacturers, yet equipment already installed will stay in use for years. Companies building devices or firmware, such as ordering kiosks, should design on the assumption that someone will try to run foreign code on them.
-
Tech
Workday and Anthropic link Sana agents with Claude
Workday and Anthropic announced a partnership connecting Workday’s Sana AI agents with Claude. Employees can start work in Claude or in Sana, exchanging context through the open Model Context Protocol. Every action follows the customer’s existing Workday permissions and audit controls, and the user confirms changes. The Sana in Claude plugin is available to early-access customers and is expected to reach general availability late in 2026.
What it means: The partnership shows how major enterprise platforms want frontier reasoning inside HR and finance workflows without loosening existing permissions. For Greek companies already on Workday or evaluating agentic tools, the key point is that rules stay in the system of record instead of drifting into an uncontrolled general chatbot. That lowers the risk of an agent approving a payment or changing pay outside company policy. It also opens practical cases such as invoice exception review or internal mobility planning inside tools teams already use. Builders of custom integrations should watch standards such as MCP and keep human confirmation for sensitive actions. Late-2026 availability leaves room for pilots, but not for delaying an agent access policy, because buyers will soon expect governed AI inside core people and money processes across everyday business operations. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs.
-
Tech
Workday to buy C1.ai for identity of people and AI agents
Workday announced a definitive agreement to acquire C1.ai, an agentic identity platform that governs access for employees and autonomous AI agents. C1.ai connects to systems such as email, CRM, code repositories, finance tools and cloud infrastructure. The deal is expected to close in Workday’s fiscal fourth quarter ending 31 January 2027. Workday frames the move as becoming the identity governance layer for people, money and agents across the enterprise.
What it means: The deal shows the market moving from experimental AI agents to access control. When agents open email, change payroll data or call payment APIs, the hard question is not only what they can do, but who granted the right and when it is revoked. Greek firms on Workday or similar HR and finance stacks should ask vendors for agent inventories, time-bounded permissions and audit trails. The same logic applies to companies building their own agents on ERP, myDATA or ordering platforms: without identity and revocation, a mistake or breach scales instantly. The move also maps to AI Act and cybersecurity expectations, where proof of control over automated systems becomes part of compliance. Software vendors selling to businesses should design role-based rights for agents now, not only for human users, because buyers will increasingly treat agent identity as a procurement requirement rather than an optional add-on in enterprise deals. It belongs in operational planning for Greek SMEs.
-
Tech
Ubitium raises $14M for a physical-AI processor
European semiconductor startup Ubitium raised $14 million in a new seed round led by Ibex Investors. It is building a reconfigurable processor that combines sensing, AI inference and real-time control for robots, drones and industrial equipment. Its first chip, the UB108, was made on Samsung’s 8nm process. The funding will advance the commercial UB410 toward production, with tape-out and evaluation units planned for 2027.
What it means: The raise shows physical AI depends not only on big data centers but also on specialised edge silicon. For Greek industry, logistics and automation firms, future robotics may get cheaper and simpler if multiple specialised chips shrink into one architecture. Product teams combining sensors and autonomous control should watch RISC-V designs and unified toolchains, because they change software cost. Foundry concentration remains a supply-chain risk. Greek system integrators can trial evaluation kits when they appear, but should avoid locking designs before 2027 hardware exists and before safety certification paths are clear for their markets. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs.
-
Tech
Nvidia pledges about $2B more for European AI investments
Nvidia CEO Jensen Huang said in Paris that the company plans to announce roughly $2 billion of European investments within a week. That would come on top of about $8 billion it has already put into the continent over the past year. He also named minority stakes in four French startups: Doctolib, Genesis AI, Pigment and Entalpic. Huang urged Europe to join the AI race more forcefully and promoted open models as a path to technological sovereignty.
What it means: Nvidia’s pledge matters for European software and fintech firms because the leading accelerator vendor is still committing capital and strategic stakes on the continent instead of concentrating only in the US and Asia. For Greek product teams, that raises the odds of nearby partners, cloud capacity and co-innovation with European startups that suddenly have Nvidia-linked backing, while GPU cost and availability remain central to roadmaps. Huang’s open-model pitch also aligns with companies that need Greek language, local data and regulatory fit without relying only on closed APIs. Builders should watch where capacity lands and which local players receive strategic money, because that shapes latency, cloud pricing and supplier choices. The cash does not rewrite the AI Act, yet it increases pressure for real productive infrastructure in the EU rather than rules alone, and Greek SMEs buying AI tooling will feel the effects through partners’ pricing and regional coverage. It belongs in operational planning, not only in a leadership news briefing.
-
Tech
NCA and Meta: nearly 30 arrests and 3.7M accounts removed in scam crackdown
The UK National Crime Agency hosted a joint operation with law enforcement and technology firms against fraud networks based in West Africa. Meta removed more than 3.7 million profiles, pages and groups. Nearly 30 people were arrested, mainly in Ghana and Nigeria, trafficking victims were rescued, and about $11 million in potentially freezable funds was identified. Participants included the FBI, Microsoft and Google among others.
What it means: The scale of removals shows platforms remain a front line for investment and impersonation scams that also hit Greek users. Businesses should train staff and customers not to trust investment pitches or urgent payments that start on social media. Brands that run ads or pages need stronger identity checks and fast reporting of lookalike pages. Public-private crackdowns will grow, so platforms may demand more verification from advertisers. Fintech and payments firms should link social-fraud signals to anti-money-laundering controls and transaction freezes before funds leave the country. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs.
-
Tech
Microsoft scales Azure Maia toward multi-gigawatt deployments
Microsoft said it will scale the Azure Maia family to multi-gigawatt deployments starting in 2027, making Maia its primary inference system for Microsoft products and services. Maia 200 is already serving production workloads including Microsoft Copilot. Maia 300 targets long-context reasoning, low-latency conversation and multi-step agents, with Microsoft claiming up to 40% more tokens per megawatt versus leading GPU systems. Production deployments of Maia 300 are planned for 2027.
What it means: Microsoft’s push into custom inference silicon changes the cost math for anyone relying on Azure Copilot, Foundry and related APIs. For Greek businesses, prices, capacity and regional availability may depend more on Microsoft’s own stack and less on third-party GPU markets. Teams building long-context or continuous agent workflows should test performance and rate limits before locking critical processes to one path. Supplier diversification remains prudent, because a strong internal platform reduces Microsoft’s incentive to price every rival model equally. Greek SaaS vendors should plan fallbacks to open or alternative models when inference cost squeezes margins, and they should revisit cloud contracts as Maia capacity ramps through 2027. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs.
-
Tech
Meta says GEM lifted ad conversions by 18.9%
Meta published a technical account of its 2026 GEM ads model, claiming an 18.9% conversion improvement and a 12% CTR lift across ad surfaces. The system unifies predictions for Facebook, Instagram and Reels in one central foundation model. The figures come from Meta itself without independent audit or a full methodology. For advertisers, the changes mostly arrive through automated delivery rather than new Ads Manager controls.
What it means: Meta’s performance claims matter for Greek businesses spending on Meta ads, because they change how the system learns from clicks and conversions across apps. Without independent measurement, the numbers should be treated as directional, not as guaranteed ROI. Advertisers should compare them with their own incrementality tests and watch cost per real sale. Cross-surface learning means behaviour on one placement can shape optimisation on another, complicating placement reports. Teams using Advantage+ should keep holdout tests and avoid raising budgets only because the dashboard shows higher conversions attributed by Meta. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs.
-
Tech
Disney bids about $20M for Character.ai IP and about $180M for staff
Character.ai filed for Chapter 11 in the United States, with Disney as the stalking-horse bidder for selected intellectual property at about $20 million. Disney also hired most of the staff, with estimated pay and retention near $180 million over three years. The deal does not buy the app, brand or user data. The platform is expected to wind down unless another buyer keeps it operating.
What it means: The case shows that in an AI startup crisis value can sit more in people and code than in the consumer app’s users. Greek AI startups should not treat an active user base as automatically liquid in bankruptcy. Buyers and investors look at talent, patents and training or safety systems, while litigation and regulation can make the whole product unattractive. Teams building consumer AI involving minors or sensitive content must invest early in compliance, because dispute costs can exceed commercial value. Business customers that rely on such tools need exit clauses and alternative vendors before a wind-down leaves them stranded. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs.
-
Tech
Anthropic asks D.C. Circuit to rehear Pentagon blacklist loss
Anthropic asked the federal appeals court in Washington to rehear the ruling that upheld the Pentagon’s supply chain risk label against the company. The September 2-1 decision left in place a bar on defense contractors using Claude. Anthropic argues the designation retaliated for its AI-safety limits on mass surveillance and lethal autonomous weapons. A parallel San Francisco case had found part of the Defense Department’s conduct unconstitutional.
What it means: The case is not a Greek SME headline, yet it sets a precedent for how easily a state can shut an AI vendor out of public and defense supply chains without a clear process. EU firms selling to government or critical infrastructure should document model-use limits, safety policies and contract clauses before a political dispute erupts. Customers that depend on a single frontier model need a fallback path, because a ban can freeze contracts overnight. The fight over permitted military uses will also shape European dual-use AI debates. Anyone embedding Claude or rival models in B2B products should treat usage terms and geographic bans as operational risk, not legal footnotes, and review concentration risk in procurement. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs.
-
Tech
Albert raises $55M for chemistry-native AI in R&D
Albert Invent announced a $55 million Series B led by TCV to expand its chemistry-native AI platform for research and development. The company highlighted its Valence model, which its research says is far more accurate than general-purpose AI at retrieving scientific answers. Enterprise customers named include BASF, Henkel and AkzoNobel. Proceeds will fund shorter product development cycles and deeper tools for scientists.
What it means: The round confirms that general large models are not enough in sectors with strict scientific vocabulary and regulatory duties. Greek chemical, cosmetics, food and materials firms should judge vertical AI by accuracy on formulations and experiments, not fluency alone. Vendors selling R&D software must offer access controls, zero data retention options and reasoning audit trails, because customers will not accept recipe leakage. Pressure for faster product-safety compliance also makes tools that cut repeated experiments attractive. SMEs can start with pilots that search internal knowledge before allowing agents to propose formulations for production, and they should involve quality and regulatory staff early. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs. It belongs in operational planning for Greek SMEs.
Summaries are ours. Copyright in the original articles remains with their publishers.
Have a project in mind?
Tell us what you need to achieve. You get a concrete proposal with scope, timeline and cost — no obligation.