TOP / Security for the AI Era

Security for the AI Era

From preparing for frontier AI threats
to crisis response after an incident.

Information Security for the AI Era

AI now finds vulnerabilities in large numbers and in a short time, and they are exploited almost at once. At the same time, the AI your company uses creates new risks of its own. DPLUS supports both fronts, defending against attackers' AI and governing your own AI, bringing technology and management together. We help you do everything you can to prevent an incident, and prepare with you, from day one, for the crisis response if defenses are breached.

Not sure where to begin? That is fine.

You do not need a defined issue to talk to us. A current-state assessment gives a view of the whole, and we decide together what to protect and in what order.

Talk to us

What Has Changed

Attacks are faster and more numerous,
and AI now takes action.

Frontier AI is good at finding vulnerabilities and writing exploit code, and attackers use it too. Meanwhile your own AI no longer just answers questions: it holds permissions and carries out work.

Found in bulk

Many vulnerabilities, in a short time

Weaknesses that used to be hard to find are now uncovered one after another by AI. The work of applying patches surges.

Exploited fast

Less time from discovery to attack

AI chains together even weaknesses with modest severity scores. Manual operations cannot keep up with the volume or the speed.

AI that acts

Your own AI acts with real permissions

Sending email, updating data, even placing orders. A wrong judgment or a malicious instruction can be carried out as is.

A security incident can stop a business overnight

Damage seen in recent incidents

  • Weeks to monthsOrders, shipments, production and services halted, at companies of every industry and size
  • Tens of billions of yenA single incident can lead to losses on this scale through lost sales and system recovery costs
  • Trust and accountabilityLate disclosure or inconsistent explanations drive away customers and partners, hit the share price and raise questions of management responsibility

Source: DPLUS, based on public disclosures and press reports from companies in Japan and overseas

Regulators treat it as a management issue

  • Led from the topLeadership allocates budget and people, and executives such as the CIO and CISO are directly involved
  • Across departmentsBusiness units, risk management, IT and security, and finance work together. One department cannot do it alone
  • Prepared for a shutdownCriteria and procedures for stopping systems, and for responding to customers, are decided in advance

Source: DPLUS, based on the METI and IPA Cybersecurity Management Guidelines Ver 3.0, and the FSA and Bank of Japan guidance on short-term responses by financial institutions to the changing threat from frontier AI (May 22, 2026), among others

Most of the work is the same. What changed is speed, volume, and what you protect.

AspectConventional securitySecurity for the AI era
Speed and volumeWork that people can handle manually, such as quarterly patchingVulnerabilities are found in bulk and exploited fast. Automation and AI become a given
Setting prioritiesFix the highest severity scores firstAI chains low-scoring weaknesses into attacks. Scores alone cannot set priorities
What you protectInternal systems, data and peopleAll of these, plus the AI you use: AI agents, AI-driven development, and the data you give AI
Role of managementOften treated as an IT department issueA management issue. Leadership owns decisions to stop systems, disclosure and crisis response

Two Fronts

Risks from outside, and risks from within.
Prepare for two fronts separately.

The risks that frontier AI brings fall into two groups: the AI attackers use, and the AI your company uses. They are different problems. Treat them separately, and prepare for both.

Front 1

Defend against attackers' AI

What happens

  • Vulnerabilities are found in bulk
  • Time to attack gets shorter
  • More people are able to attack

How to prepare

  • Do the basics, fast
  • Fight back with AI
  • Prepare to stop and recover
  • Be ready for the crisis that follows
More on Front 1

Front 2

Govern your own AI

What happens

  • Confidential data leaks through AI
  • AI agents make mistakes, or are hijacked
  • AI-written code leaves weaknesses behind

How to prepare

  • Set usage rules and ownership
  • Classify and control data
  • Design permissions and human oversight
  • Verify that controls actually work
More on Front 2

Technology alone cannot protect you. What it takes is the ability to bring it all together.

Patch management, vulnerability assessments, multi-factor authentication, EDR and monitoring, backups. Each measure can be bought from a product or a vendor. Only when governance, overall design and execution bring them together are you actually protected.

Governance

Who decides, and who is accountable

We define the roles of management, business units, IT, legal and communications, and the criteria for decisions.

Overall design

What to do, in what order, and how far

We set priorities and allocate investment based on risk and business impact.

Execution (PM)

Seeing it through across departments and vendors

We make progress and gaps visible, get the work done by the deadline, and report to leadership.

Front 1

Front 1: Defend against attackers' AI

Before adding new measures, we build the capability to do the basics quickly and reliably. Then, assuming not every attack can be stopped, we prepare for decisions to stop systems, recovery, and what comes after.

ST0・ST1

Do the basics, fast

Technology assessment / Faster, automated vulnerability response

Challenges

  • Vulnerabilities pile up faster than they can be fixed
  • Severity scores alone cannot set priorities
  • Old equipment and unused connections leave too much to protect

Direction

  • Identify critical services and systems, and start there
  • Prioritize by how likely an attack is to succeed
  • Reduce technical debt (unused ports, privileged IDs, unsupported products)
  • Check maintenance vendors' contracts and capabilities

How DPLUS helps

  • Governance, technology, and people and organization assessments
  • Rules for setting priorities
  • Response plan and structure, and driving execution
  • Review of vendor contracts

Example deliverables

  • Current-state assessment report
  • List of priority systems
  • Vulnerability response policy and procedures
  • Improvement roadmap
Manufacturing

A common situationThree months after a critical vulnerability was announced, the factory's VPN devices were still unpatched. Fixes were made in order of severity score, and no one had decided which devices mattered most.

With our supportCritical systems are identified first, and weaknesses that could actually be exploited are fixed first. A setup is put in place to address critical weaknesses within days, with matching agreements with the maintenance vendor.

ST2・ST3

Fight back with AI

AI-powered defense / SOC (monitoring and detection)

Challenges

  • Attacks run at machine speed; defense is still manual
  • Too many alerts bury the important ones
  • AI is adopted piecemeal and never becomes part of operations

Direction

  • Decide which parts of the detect-to-fix flow AI should handle
  • Automate prioritization, fix proposals and routine responses first
  • Keep human approval for important decisions
  • Combine AI with layered defenses rather than relying on it alone

How DPLUS helps

  • Policy for using AI and the work it covers
  • Support in selecting tools and specialist firms
  • Operating rules and human oversight
  • Embedding it in daily operations

Example deliverables

  • Policy for AI in defense
  • Automation scope and procedures
  • Implementation plan
  • Operating rules
Retail and e-commerce

A common situationTwo monitoring staff received thousands of alerts a day. Signs of an intrusion sat among the alerts they had put off, and it took days to discover.

With our supportAI handles first-pass triage and routine responses, so people can focus on judging the important alerts. High-impact actions, such as isolating a system, run only after a person approves them.

SG2・SG3

Prepare to stop and recover

Business continuity and shutdown decisions. Stopping is a management decision, so set the criteria in normal times

Challenges

  • Even thorough measures cannot stop every attack
  • No criteria or decision-maker for stopping systems
  • The business continuity plan does not assume a cyberattack

Direction

  • Set criteria and procedures for a deliberate shutdown
  • Review the business continuity plan and emergency contacts
  • Prepare procedures for responding to customers and others
  • Train, including leadership

How DPLUS helps

  • Criteria for shutdown decisions
  • Review and revision of the business continuity plan
  • Incident response structure
  • Tabletop exercises including leadership

Example deliverables

  • Shutdown criteria and procedures
  • Revised business continuity plan
  • Incident response procedures
  • Exercise results and improvements
Logistics

A common situationRansomware was detected late at night. While staff waited for an executive to decide whether to stop the systems, the damage spread to other sites.

With our supportClear criteria ("with these signs, this person stops the systems") and a backup decision-maker are set, and leadership trains on them. Even at night, systems can be stopped right away by the agreed procedure.

Crisis Management

Crisis management (SH3, SH6–SH8)
Technical response and accountability, as one.

After an incident, you must explain what happened to the outside world while the technical response is still under way. Get the response wrong and the damage keeps growing. We build everything, from preparation in normal times to disclosure, press conferences and rebuilding trust, as a single flow.

Prepare in normal times

  • Crisis communications manual, structure and contact tree
  • Criteria for disclosure, and what can and cannot be said
  • Media training and mock press conferences
  • Ongoing social media monitoring

Right after an incident

  • Setting up a response headquarters
  • Establishing the facts with the technical investigation
  • Forecasting the impact, and tracking press and social media
  • Deciding whether, when and to whom to disclose
  • First reports to regulators and business partners

Disclosure and briefings

  • Drafting and editing press releases and timely disclosures
  • Running emergency press conferences and the press room
  • Advising spokespeople
  • Setting up an inquiry desk
  • Explaining to customers, shareholders and partners

Containment and trust

  • Monitoring press and social media, preventing a second wave
  • Countering harmful rumors, including in search results
  • Compiling and publishing recurrence prevention measures
  • Coordinating with third-party committees and others

Consistent explanations for every audience, from one team

The press, shareholders and investors, customers, regulators, business partners, employees, social media. We bring the explanations for audiences in very different positions into one plan. We team up with lawyers and specialists in crisis communications and social media, and DPLUS leads the whole effort, through to preparing and running press conferences.

Front 2

Front 2: Govern your own AI

Governance is not a brake on using AI. It is what lets you hand more work to AI with confidence. Before technical measures, we decide who hands what to AI, and how far.

G1

AI Usage Policy

Rules that widen what you can safely hand to AI

Challenges

  • No rules, so staff decide for themselves
  • Or so many bans that adoption stalls
  • No one is clearly in charge

Direction

  • Decide which data and uses are allowed
  • Set up a structure and a person in charge
  • Assess risk by use case
  • Communicate the rules and train people

How DPLUS helps

  • Understanding how AI is actually used
  • Drafting the AI usage policy
  • Designing the structure and roles
  • Internal briefings and training

Example deliverables

  • AI usage policy
  • Usage guidelines
  • Structure and role chart
  • Training materials
Services

A common situationAfraid of data leaks, the company banned generative AI outright. Employees started using personal accounts for work, and even entered customer information.

With our supportAllowed data and uses are set out in tiers, and company-approved AI is provided. Hidden use goes down while the work people can safely do with AI grows.

G2

AI Security & Data Governance

We check not only where data is stored, but whose control it is under

Challenges

  • No criteria for which data can go into which AI
  • Unclear which country processes the data and who controls it
  • Contracts with AI vendors have not been checked

Direction

  • Classify data by sensitivity and decide which AI can use it
  • Confirm where data is stored, processed and logged
  • Set contract terms such as no use for training and retention periods
  • Inspect inputs before and outputs after

How DPLUS helps

  • Data classification and input rules
  • Evaluating AI vendors
  • Items to confirm in contracts
  • Building it into operations

Example deliverables

  • Data classification criteria
  • Allowed-use table by AI service
  • Vendor evaluation sheet
  • Contract checklist
Financial services

A common situationA business unit signed up for a convenient overseas AI service on its own. Only later did it learn that the contract allowed input data to be used for training and stored overseas.

With our supportData is classified by sensitivity, and an allowed-use list is kept for each AI service. Before signing, a step checks use for training, storage location and retention period.

ST4

AI Agent Permissions & Human Oversight

Treat an AI agent as a separate worker acting on a person's behalf

Match human oversight to the risk

Type of actionExamplesHuman oversight
Read and summarizeSummarizing and sorting internal documentsRuns automatically
Update internal dataUpdating ledgers and recordsReviewed afterwards
Send externallyEmails to customers, handling inquiriesApproved beforehand
Money, permissions, deletionPayments, permission changes, deleting dataTwo-person approval

Permission principle: give each agent its own ID, with only the minimum permissions it needs, for a short period

Challenges

  • AI runs under a person's account, so no one can tell who did what
  • Requiring approval for everything brings work to a halt

How DPLUS helps

  • Assessing risk by task, and deciding what to hand over
  • Designing approvals (what, by whom, on which screen)
  • Requirements for permissions and logs

Example deliverables

  • Permission and approval table by task
  • Operating rules for agents
ST5

Security for AI-driven Development

AI-written code gets the same checks. People remain accountable

AspectWhat happensDirection
Quality of generated codeTests pass, but weaknesses such as loose authentication remainAlways run the same checks as for code written by people
Suggested packages that do not existAI suggests a library name that does not exist, and an attacker publishes a malicious package under that name (slopsquatting)Confirm a package exists and where it comes from before adding it, and allow only approved packages
Prompt injectionAI follows instructions hidden in external documents, web pages or code commentsAssume anything AI reads may contain instructions
Permissions and secretsAI reaches credentials and configuration files beyond what it needsLimit what it can read, and never give it production permissions
OverconfidenceDevelopers accept AI output without checking itName who is responsible for review and set approval points

How DPLUS helps: development rules / building checks into the process / taking stock of AI agent permissions / review responsibility and approval points

G5

AI Audit & Responsible AI

We check that controls actually work, not just that they exist on paper

Challenges

  • Rules exist, but no one checks they are followed
  • No answer when asked to explain an AI decision
  • Unable to keep up with new regulations and guidelines

Direction

  • Check regularly that controls are working
  • Keep records of AI use so it can be traced
  • Run tests that simulate attacks
  • Check against regulations and guidelines in Japan and abroad

How DPLUS helps

  • Audit criteria and procedures
  • Running reviews and proposing improvements
  • Checking against regulations and guidelines
  • A reporting process for leadership

Example deliverables

  • AI audit checklist
  • Review report
  • Improvement plan
  • Leadership report template
Insurance

A common situationThe AI usage rules had not been touched since they were written two years earlier. When a major client asked how AI was being used, there were no records to show, and no answer.

With our supportRegular reviews and checks of usage records become routine. Unapproved AI tools are found and sorted out, so the company can explain actual use to clients and regulators.

Example frameworks: Japan's AI Guidelines for Business, the EU AI Act, the NIST AI Risk Management Framework, ISO/IEC 42001

How to Start

Start with an assessment,
set priorities, and move in stages.

You can engage us for selected services only. We start by looking at the whole with a current-state assessment, then decide together what support you need.

Assessment

Choose from governance, technology, and people and organization assessments, and identify what to protect first

Policy and priorities

Agree with leadership on the order of measures and who owns them, and build a roadmap

Execution

Put rules, structures and procedures in place. Technical measures go ahead with specialist engineers, and are tested with fire-drill style simulations

Embed and review

Make it part of daily work and review it regularly, adjusting to changes in threats and regulation

Our strength: from technology to management, in one team

01

We start by defining the issue

Not from a list of technical measures, but by deciding with leadership what to protect and how far to trust AI.

02

We see governance and execution through

We set the overall design and priorities, and act as PM across departments and vendors until the work is done.

03

Technology and management, together

We team up with vulnerability assessment and forensics engineers, former police officers, lawyers and people experienced in press response, and DPLUS stays accountable to the end.

04

We support you after an incident

From the technical first response to IR, press relations, press conferences and customer response, we lead the whole crisis response.

In M&A, we also arrange security due diligence to check the real security posture of a target. See Corporate Transformation.

Self Check

Self-check: 10 questions

Areas with many "no" or "not sure" answers are the first candidates to tackle.

Defending against attackers' AI

  1. Can you quickly identify your critical systems and how they are built?
  2. Do you prioritize vulnerabilities by criteria other than scores?
  3. Are there unsupported products or unused connections left?
  4. Are there criteria and a decision-maker for stopping systems?
  5. Are there procedures for disclosure and customer response after an incident?

Governing your own AI

  1. Have you decided which data and uses are allowed with AI?
  2. Do you know where data is stored and processed, and on what contract terms?
  3. Do AI agents have their own IDs and minimum permissions?
  4. Is there human approval for important actions?
  5. Is AI-written code checked to the same standard as code written by people?

Frequently asked questions

Do we have to buy a particular security product?

No. DPLUS does not sell products. Once we have decided what to protect and in what order, we help you choose suitable tools and specialist firms, including the products you already use.

Can we ask for selected services only?

Yes. The service map is there to share the whole picture and the order of work. You can start, for example, with crisis preparation in normal times, or with an AI usage policy alone.

Can we come to you after an incident has happened?

Please do. Alongside the technical first response, we team up with specialists to support disclosure, press relations, press conferences and customer response.

Can you also help us use AI itself?

Yes. Adopting AI, spreading its use and redesigning work are covered by AI Shift. Governance and adoption can also be run together.

Contact

What to protect, and where to start.
Let's decide together.

Talk to us even before the issue is clearly defined.