Coverage period: stories published or updated during the last 24 hours.

Review base: 312 submitted articles from 33 publications and organisations. Related reports have been consolidated into five distinct business developments.

1 · Impact score 10/10 · Immediate UK relevance

UK airport hackers publish data on 8.7 million travellers

Manchester Airports Group, which operates Manchester, Stansted and East Midlands airports, has reportedly had stolen customer data published after refusing to pay a ransom.

Why it matters: This is a high-profile reminder that a cyber incident can become a long-term privacy, fraud and reputational problem even when the organisation does not pay a ransom. UK businesses holding customer, employee or identity data face similar exposure.

Commercial implications: Leaked data can drive phishing, impersonation, account takeover and compensation claims. Customers and business partners may also demand stronger evidence of cyber resilience during procurement and contract renewals.

Sectors affected: All sectors, particularly travel, hospitality, retail, financial services, healthcare, education and professional services.

Risks and opportunities: Risks include regulatory scrutiny, customer loss, operational disruption and increased insurance costs. The opportunity is to use this incident to strengthen data minimisation, supplier assurance and incident response before a breach occurs.

Next steps:

• List the personal and commercially sensitive data your business holds, where it is stored and who can access it.
• Check that your incident plan includes customer communications, regulator notifications and supplier coordination.
• Test backups, privileged-account controls and phishing resilience.
• Ask key technology suppliers how they would notify you and contain a breach affecting your data.

Further reading: BBC News · Computer Weekly · TechRadar Pro

2 · Impact score 9/10 · Cybersecurity and governance

AI agents are beginning to automate the full cyber-attack chain

Several reports describe AI systems finding vulnerabilities, progressing through ransomware attacks and producing detailed post-attack reports.

Why it matters: The threat is not simply that AI makes existing attacks faster. Automated systems can now scan, exploit, move through networks and perform repetitive attack tasks with limited human involvement. This lowers the cost of attacking smaller organisations.

Commercial implications: Businesses should expect more convincing phishing, faster vulnerability exploitation and less warning between compromise and disruption. AI adoption also creates a second risk: poorly controlled internal agents may make unauthorised changes or expose information.

Sectors affected: All sectors; especially manufacturing, energy, transport, healthcare, financial services, software and organisations operating older systems.

Risks and opportunities: Risks include ransomware, service outages, data theft and unsafe automation. Defensive AI tools may help smaller teams identify weaknesses more quickly, but they must support rather than replace human judgement.

Next steps:

• Identify every AI tool or agent currently used by staff, including personal accounts and free trials.
• Give agents the minimum access needed and keep high-impact actions behind human approval.
• Prioritise patching internet-facing systems, remote access tools and old infrastructure.
• Run a tabletop exercise covering ransomware, supplier compromise and an AI-assisted attack.
• Make sure your managed service provider can monitor unusual automated activity.

Further reading: The Register · TechRadar Pro · Computer Weekly · UK cyber policy context

3 · Impact score 9/10 · Immediate security action

Stolen AI-service sessions can reach connected corporate accounts

Infostealer malware has reportedly been used to steal Claude login sessions. Replaying a stolen session can bypass the login page, including two-factor authentication.

Why it matters: Many employees use AI services through personal subscriptions on work devices. If those accounts are connected to Gmail, Microsoft 365, cloud storage or other business systems, an attacker may gain access without the business knowing the account exists.

Commercial implications: The direct cost of stolen AI usage may be small compared with the risk of exposed emails, files, customer information and connected applications. Traditional single sign-on controls do not protect accounts that employees create and manage themselves.

Sectors affected: All sectors, particularly professional services, technology, financial services, legal, recruitment and marketing.

Risks and opportunities: Risks include data theft, fraudulent messages, unauthorised payments and loss of confidentiality. The opportunity is to bring AI use into normal identity, device and data-governance controls.

Next steps:

• Ask staff to declare AI accounts used on company devices or with company data.
• Block personal AI accounts from connecting to corporate email, storage and customer systems where possible.
• Move frequent users to an organisation-managed AI plan with administrator controls.
• After a malware alert, revoke AI sessions and connected application permissions, not just the device password.
• Remind staff to download AI software only from verified sources and to avoid pirated software.

Further reading: VentureBeat · Malwarebytes · Anthropic guidance on Workspace connectors

4 · Impact score 8/10 · AI capability and cost

Google launches Gemini 3.8 Flash and a cyber-defence variant

Google says the new model is designed for multi-step work and AI agents, while Gemini 3.8 Flash Cyber is focused on finding and fixing software vulnerabilities.

Why it matters: Frequent improvements in lower-cost AI models are making useful automation more accessible to smaller businesses. The latest release also shows that cyber defence is becoming a practical application for advanced AI, rather than a future research topic.

Commercial implications: Potential uses include document processing, customer-service support, internal research, software development and security checks. However, model pricing is only one part of the cost: review, integration, data protection and error handling still matter.

Sectors affected: All sectors; especially software, professional services, finance, retail, logistics and customer-service operations.

Risks and opportunities: Opportunities include faster handling of repetitive knowledge work and cheaper experimentation. Risks include incorrect outputs, rising usage bills, supplier dependency and sensitive data being sent to an external service.

Next steps:

• Choose one low-risk, repetitive process for a controlled trial rather than rolling out AI broadly.
• Measure time saved, error rates, review time and total cost against the current process.
• Set a monthly usage limit and require approval before connecting the model to business systems.
• Check data-processing terms, retention settings and UK GDPR responsibilities.

Further reading: Google announcement · VentureBeat analysis · Ars Technica

5 · Impact score 7/10 · Accessible data and analytics

Google adds predictive analytics to BigQuery without model training

BigQuery’s preview TabFM feature lets users generate classification and forecasting predictions from existing structured data using SQL.

Why it matters: Businesses already using Google Cloud may be able to test useful predictions without hiring specialist data scientists or creating a separate machine-learning platform. Possible examples include customer churn, fraud indicators, demand and future claim values.

Commercial implications: This could shorten the route from business data to practical insight. It is most suitable for experimentation and lower-volume use cases, not necessarily as a replacement for established models handling large or high-frequency workloads.

Sectors affected: Retail, professional services, finance, insurance, healthcare, logistics, manufacturing and subscription businesses.

Risks and opportunities: Opportunities include better targeting, earlier intervention and more informed planning. Risks include biased or inaccurate predictions, unclear explanations and costs increasing when token-based pricing begins on 30 October 2026.

Next steps:

• Check whether your existing data is clean, consistent and labelled with known outcomes.
• Pick a contained use case where a wrong prediction will not cause serious harm.
• Compare predictions with historical results and require a human review before action.
• Model expected usage costs before moving beyond a pilot, including BigQuery charges.
• Do not use predictions for sensitive decisions without checking fairness, explainability and legal requirements.

Further reading: InfoWorld · Google Cloud announcement · BigQuery AI.PREDICT documentation