Video summary
SoftBank World 2026 宮川潤一 特別講演
Main summary
Key takeaways
Executive summary
SoftBank CEO Junichi Miyakawa argued that AI is changing both the operating model of businesses and the threat landscape. Companies should not treat cybersecurity as a matter of restoring systems after an outage; they should modernize systems so they are more resilient and can remain operational under attack.
His central recommendation was to modernize corporate systems for the AI era: use AI to find and fix vulnerabilities while rebuilding legacy systems and operating practices designed around human-led development and maintenance. Because companies’ systems are interconnected through suppliers and critical infrastructure, he framed cybersecurity as a shared responsibility—not just an individual company’s concern.
Business strategy and operating lessons
- Build resilience into the foundation. Miyakawa compared system modernization to Chicago’s post-fire rebuilding: rather than simply recreate what had burned, the city changed its underlying design and building standards. Companies, he argued, should likewise rebuild vulnerable system foundations instead of relying only on recovery plans.
- Treat cybersecurity as part of business continuity and sustainability. A cyberattack can disrupt operations and reduce sales, but the more serious damage may be loss of customer trust and risk to the company’s ongoing viability.
- Move from human-centered to AI-centric development and operations. Systems assembled by multiple teams and vendors can accumulate inconsistent settings, gaps, and implementation errors. AI may identify connections between those weaknesses that are difficult to detect through manual review.
- Account for supply-chain and infrastructure effects. A company’s defenses can be undermined through connected vendors or shared systems. Miyakawa urged companies to strengthen their own defenses while recognizing that broader protection requires cooperation.
- Use AI to improve service, not merely automate existing processes. In customer support, he emphasized adapting workflows to customer feedback and organizational practices rather than treating automation as a one-time replacement of human labor.
Frameworks and playbooks
AI-era cybersecurity modernization
Miyakawa outlined a four-step vulnerability and remediation workflow:
- Identify vulnerabilities and propose fixes with AI. He said this stage had already been fully automated in SoftBank’s process.
- Test proposed changes before deployment. Automated testing was in development so companies could assess the likely results of applying a patch.
- Check system interconnections. Validate that a fix does not cause problems in neighboring systems. This stage was still partly manual, with full automation targeted within roughly a month.
- Apply patches and verify results. Check for bugs and similar vulnerabilities, and determine whether a remediation can be rolled out to other systems. Miyakawa estimated this broader capability would take another two to three months.
A key governance principle: Customers retain the decision to proceed at each stage. SoftBank’s role is to provide diagnosis and tools that make decisions easier, not to assume responsibility for the customer’s service.
Immediate response recommended to companies
- Start with the systems of greatest concern rather than attempting to assess everything at once.
- Identify weaknesses, prioritize them, and apply patches quickly.
- Treat patching as preventive protection—likened by Miyakawa to a vaccine—rather than waiting for an attack and then trying to recover.
- Assign a company contact and select the systems to be assessed as initial steps for engaging with SoftBank’s diagnostic offering.
Products, sales, and customer-service examples
- Japanese-language AI translation: Miyakawa demonstrated a translation model running through a data center in Saitama, reporting about 0.5 seconds of latency. He said it was a lighter, task-specific version of a larger model, tuned for Japanese wording and pronunciation. The demonstration included voice conversion and speech in the speaker’s voice; he said the service supported eight languages.
- AI contact-center tools: SoftBank showed an updated version of its Genax/CrossCoast offering, which can turn existing manuals and call flows into procedures and incorporate frontline knowledge through dialogue. The pitch was that workflows can be updated to suit a customer’s operations and culture.
- Sierra collaboration: SoftBank announced a collaboration with Sierra to expand its AI contact-center business in Japan. Miyakawa highlighted implementation in two weeks, describing a process based on an API connection and discussions with a lead engineer about the intended scenario.
- Illustrative telecom support interaction: An AI agent explained a customer’s higher bill, verified identity by SMS, asked about usage, recommended a lower-cost plan, and sent a link to change it. In the example, the customer’s monthly plan moved from ¥6,930 to ¥4,180, a stated saving of ¥2,750 per month. Miyakawa used the example to show how an agent can handle both customer conversation and actions across systems.
Cybersecurity evidence and metrics cited
Miyakawa presented the following figures from SoftBank’s assessments and demonstrations:
- SoftBank had approximately 1,800 systems, of which about 700 contained internally developed source code.
- A test of those 700 systems found 10,500 vulnerabilities, or about 15 per system on average, though vulnerabilities were unevenly distributed.
- SoftBank said it had received diagnostic requests from 137 companies and completed assessments for about 50.
- For commercial systems standardized as having at least 10 million lines of code, the reported average was 280 vulnerabilities.
- In an authorized, non-production e-commerce test, an AI cyber model reportedly breached a site in 1 hour 57 minutes, despite a prior assessment that found no high-risk vulnerabilities and only four medium- or low-risk ones.
- A second test system, with three high-risk vulnerabilities, was reportedly breached in 11 minutes.
- Miyakawa said a repeat test using a “mosaic” approach to change the system’s structure ended in failure for the attack, illustrating the potential value of redesigning systems rather than relying only on perimeter defenses.
These were presented as examples and company test results, not as universal benchmarks.
AI-related outlook
Miyakawa said AI was progressing from text tools to systems that can process multiple forms of information, act as agents, and operate in the physical world through robotics. He warned that cyber-capable AI could lower the barrier to attack, operate continuously, and pursue multiple attack paths in parallel. He also forecast that more accessible cyber models could emerge in the coming months; this was his outlook, not a confirmed timeline.
Presenters and sources
- Junichi Miyakawa, President, CEO and Representative Director, SoftBank Corp. — keynote presenter.
- Miyakawa referenced Masayoshi Son and Murakami in discussing AI and cybersecurity.
- A Sierra representative delivered a prerecorded partnership message; the speaker’s name is unclear in the subtitles.
- Company examples and figures were presented by Miyakawa on behalf of SoftBank.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.