Tech and AI this week: new models, cyber warnings, local moves

This week's technology stories fall into three themes: new AI models from large companies, the security risks that come with AI, and developments inside Pakistan. The stories are linked, because the same tools that are becoming cheaper and faster are also the ones that officials warn can be used for attacks.

The first theme is new models. khabr.pk reported on 10 November that Google has begun rolling out Gemini 4 Carbon to the public, with introductory pricing of $2 per million input tokens and $10 per million output tokens, doubling at the end of November. On the same day, Anthropic released Claude Fable 5.2 and Mythos 5.2. It says outputs are 30% faster and the models cost 20% less than the 5.1 generation. On 11 November, OpenAI added verified Business Agents to ChatGPT, which let users chat with an AI representing a brand. That beta is open first to eligible US businesses. For Pakistani businesses and developers who use these services, falling prices and new features matter, though some of these offers are limited to specific groups or countries.

The second theme is security. On 6 November, khabr.pk reported that OpenAI launched GPT-6 Cyber, a model for finding vulnerabilities. OpenAI says it went out of bounds less than 1 percent of the time in testing, down from around 5 percent for its predecessor. The same week, Pakistan's National CERT issued an advisory, dated 5 November, warning that AI and automated tools are being used for coordinated cyberattacks. It named banks, oil and gas, water, power and transport systems as targets. It advised banks to use multi-factor authentication and to check customer-facing balances against internal records, and told operators to audit network access and isolate critical systems.

The link between these two stories is direct. Security models are built to find weaknesses, while the advisory warns that attackers use similar automation. For households, the risk is to bank accounts and payment apps. For businesses, it is to the systems that keep services running.

The third theme is Pakistan's own technology sector. On 6 November, Planning Minister Ahsan Iqbal said the National AI Policy aims for one million AI professionals by 2030, with seven new AI centers linking research, skills training, startups and industry. On 9 November, khabr.pk reported that a Pakistani engineer built two robot waitresses, Robia and Jojo, now serving at a Karachi restaurant. On 7 November, it reported that six satellite internet companies have registered with the SECP, but registration alone does not permit service. Starlink, Galaxy Space and SpaceCell Pakistan await Ministry of Interior security clearance, and firms then need a PSARB NOC and a PTA licence.

Questions about young users also ran through the week. Meta is expanding AI age detection on Instagram to all regions to place suspected teenagers in Teen Accounts. Google will keep Gemini and AI-linked Search off by default for school accounts until administrators switch them on. The European Commission preliminarily found that TikTok's For You feed breaches the Digital Services Act, which TikTok rejects.

Much remains open. Google has set no date for its school rollout. Satellite internet still needs several approvals. The reports do not say when the new AI features will reach Pakistani users.