AI and Cybersecurity Explained: How AI Is Changing Attacks and Defense

AI and cybersecurity are becoming increasingly difficult to separate. Artificial intelligence can help security teams analyze enormous volumes of alerts, discover vulnerabilities, investigate suspicious activity and respond to incidents faster. The same capabilities can also help attackers research targets, improve phishing messages, develop malicious tools, search for vulnerabilities and automate parts of an attack.

The important point is not that AI has suddenly invented an entirely new form of cybercrime.

Current research shows something more complicated.

Attackers are mostly integrating artificial intelligence into familiar cyberattack techniques and making parts of those workflows faster, cheaper or easier to scale. At the same time, cybersecurity teams are using AI to process volumes of security information that would be extremely difficult for humans to examine manually.

Verizon’s 2026 Data Breach Investigations Report says threat actors are using generative AI across multiple stages of attacks. In its analysis, the median researched threat actor sought AI assistance across around 15 different MITRE ATT&CK techniques, while some actors used AI across 40 or 50 techniques.

But Verizon also found that most AI-assisted malware activity was still associated with well-known attack methods rather than completely novel techniques.

That distinction is essential.

AI is already changing cybersecurity, but much of its immediate impact comes from accelerating existing attacks rather than replacing them with science-fiction-style autonomous hacking.

Google Threat Intelligence Group reached a similar conclusion, although its May 2026 research documented an important escalation: for the first time, Google said it had identified a threat actor using a zero-day exploit that it believed had been developed with AI assistance.

For the broader explanation of phishing, ransomware, malware, identity attacks, data breaches and cyber resilience, start with The News Ink’s Cybersecurity Explained guide.

AI and Cybersecurity at a Glance

Cybersecurity area	How attackers can use AI	How defenders can use AI
Reconnaissance	Research targets and technologies	Analyze exposure and attack surface
Phishing	Generate personalized messages	Detect suspicious language and behavior
Vulnerability research	Search for software weaknesses	Find and prioritize vulnerabilities
Malware	Assist coding and debugging	Analyze malicious code
Identity attacks	Create synthetic identities and impersonations	Detect anomalous authentication
Threat intelligence	Research defensive technologies	Process large volumes of threat data
Incident response	Adapt attack techniques faster	Summarize and investigate incidents
Deepfakes	Impersonate executives or relatives	Detect suspicious media and behavior
AI agents	Potentially automate attack steps	Contain compromised accounts automatically
AI systems	Prompt injection and model attacks	Red teaming and AI-security monitoring

This dual-use nature explains why AI and cybersecurity cannot be understood only from the attacker’s perspective.

The same underlying capability can serve both sides.

What Does AI and Cybersecurity Actually Mean?

AI and cybersecurity covers three related but separate problems.

AI Used to Improve Cyber Defense

Security teams can use artificial intelligence for activities such as:

• alert analysis;
• malware investigation;
• anomaly detection;
• threat intelligence;
• vulnerability prioritization;
• code review;
• incident summarization;
• and automated response.

AI Used to Assist Cyberattacks

Threat actors can use AI for:

• reconnaissance;
• phishing;
• translation;
• scripting;
• vulnerability research;
• malicious code development;
• impersonation;
• and attack automation.

Protecting AI Systems Themselves

AI applications introduce their own attack surface.

Organizations must think about:

• prompt injection;
• poisoned data;
• sensitive-data leakage;
• model manipulation;
• excessive agent permissions;
• insecure plugins or tools;
• compromised AI supply chains;
• and unauthorized model access.

NIST’s emerging Cyber AI Profile reflects these three areas almost exactly.

Its framework organizes AI cybersecurity around:

Secure: Protect AI systems.

Defend: Use AI to improve cybersecurity.

Thwart: Reduce AI-enabled cyberattacks.

NIST Cyber AI Profile

That is a useful framework because organizations need to consider all three simultaneously.

AI Is Making Reconnaissance Faster

AI and cybersecurity are becoming increasingly difficult to separate. Before attacking a company, criminals need information.

They may research:

• employees;
• technologies;
• software versions;
• cloud services;
• suppliers;
• public documents;
• exposed infrastructure;
• job advertisements;
• and organizational relationships.

Generative AI can accelerate this research.

Instead of manually reading dozens of documents, an attacker can use AI to summarize material, identify technologies and organize information.

Google Threat Intelligence Group reported in February 2026 that threat actors were increasingly integrating AI into reconnaissance, social engineering and malware-development workflows.

This does not necessarily give an inexperienced criminal instant expert-level hacking ability.

It reduces the cost of gathering and organizing information.

That matters because reconnaissance is usually repetitive.

AI performs repetitive information-processing work extremely well.

AI Can Make Phishing More Convincing

Phishing remains one of the clearest areas where AI can improve an existing attack.

Older phishing campaigns frequently contained:

• awkward language;
• spelling mistakes;
• poor translation;
• generic greetings;
• obviously fake formatting.

Generative AI can remove many of those weaknesses.

Attackers can create messages tailored to a person’s:

• job;
• company;
• language;
• location;
• industry;
• or current business activity.

Microsoft’s Digital Defense Report says attackers are already using AI to automate phishing and produce more convincing synthetic content.

Verizon’s 2026 research also found that mobile-focused social-engineering simulations produced click rates 40% higher than traditional email phishing.

That means security education needs to evolve.

Employees can no longer be told simply:

“Look for bad grammar.”

A perfectly written message can still be malicious.

The better defense is verifying the request.

The News Ink’s phishing warning signs guide explains why links, identity, urgency and requested actions now matter more than grammar alone.

AI Can Scale Social Engineering

AI and cybersecurity are becoming increasingly difficult to separate. Cybersecurity is not only a technical battle.

Attackers frequently manipulate people.

Generative AI can make social engineering more scalable because one criminal can produce many convincing variations of the same attack.

A scam operation could create:

• different personas;
• translated messages;
• follow-up replies;
• fake customer-service conversations;
• fake job offers;
• fraudulent investment pitches;
• or impersonation scripts.

OpenAI’s February 2026 threat research found that malicious actors typically used AI as one component in larger operations involving traditional websites, social-media accounts and other infrastructure.

That finding is important because it challenges one common misconception.

AI usually does not conduct the entire crime by itself.

It becomes another tool inside a broader criminal operation.

Deepfakes Create an Identity Problem

AI-generated voice and video create a different cybersecurity challenge.

A criminal can potentially imitate:

• a chief executive;
• finance manager;
• colleague;
• family member;
• government official;
• or customer.

The objective may be to persuade someone to:

• transfer money;
• disclose information;
• reset a password;
• approve MFA;
• or grant access.

Microsoft’s latest defense research says AI-generated identity forgeries are becoming significantly more sophisticated, including synthetic documents and media designed to defeat verification processes.

This weakens an assumption that humans have relied on for decades:

Seeing or hearing someone is proof that the person is really there.

It no longer is.

For important requests, organizations should verify identity through another trusted channel.

The News Ink’s deepfake scams guide explains why modern fraud prevention needs independent verification instead of relying only on whether a voice or video appears authentic.

AI Can Assist Malware Development

One of the most heavily debated questions around AI and cybersecurity is whether generative models can help create malware.

The answer is yes, but context matters.

AI coding systems can help with legitimate programming tasks such as:

• writing functions;
• debugging errors;
• translating between programming languages;
• explaining APIs;
• or modifying existing code.

Those same capabilities are dual use.

OpenAI has previously reported disrupting threat actors that used AI to help develop or refine malware, troubleshoot malicious code and configure command-and-control infrastructure.

Google’s threat research has observed similar use of generative AI for scripting and malware-development support.

But current evidence should not be exaggerated.

Verizon’s 2026 DBIR found that the overwhelming majority of AI-assisted malware observations involved attack techniques for which many existing malware examples already existed.

Less than 2.5% of the AI-assisted malware observations in Verizon’s analysis involved particularly uncommon techniques with one or fewer known malware examples.

So the main short-term concern is not:

“AI has invented completely new malware nobody has ever seen.”

It is:

“AI can make existing malicious-development work faster and accessible to more people.”

AI Is Becoming More Important in Vulnerability Discovery

Vulnerability research may eventually become one of the most consequential areas of AI and cybersecurity.

Modern software contains enormous amounts of code.

Security researchers search that code for mistakes that attackers could exploit.

AI can assist by:

• reviewing source code;
• identifying suspicious patterns;
• comparing code with known vulnerabilities;
• generating tests;
• analyzing crashes;
• and reasoning about possible exploitation paths.

That capability benefits defenders because vulnerabilities can be discovered before attackers exploit them.

But it creates a race.

Google Threat Intelligence Group reported in May 2026 that it had identified what it believed was the first observed case of a threat actor using a zero-day exploit developed with AI assistance.

Google’s May 2026 AI Threat Tracker

A zero-day vulnerability is particularly dangerous because defenders may not know the weakness exists when exploitation begins.

This development makes patching speed even more important.

Verizon’s 2026 DBIR already says vulnerability exploitation has become the leading breach entry point, accounting for 31% of breaches in its dataset.

If AI reduces the time needed to identify or weaponize weaknesses, defenders may have less time to patch.

AI Does Not Eliminate the Need for Cybersecurity Fundamentals

AI and cybersecurity are becoming increasingly difficult to separate. The growth of AI-enabled attacks can create the impression that every company needs an equally futuristic defense system.

That would be a mistake.

AI does not make basic cybersecurity obsolete.

It makes basic cybersecurity more urgent.

Organizations still need:

• timely patching;
• MFA;
• phishing-resistant authentication;
• least privilege;
• endpoint security;
• secure backups;
• network segmentation;
• logging;
• incident-response planning;
• and vendor security.

An AI-assisted phishing message still becomes far less useful if stolen passwords cannot bypass MFA.

AI-assisted vulnerability research matters less against systems that have already been patched.

AI-generated malware becomes less damaging when endpoint controls, restricted privileges and network segmentation limit its execution.

This is why the main The News Ink Cybersecurity Explained pillar remains the foundation beneath AI-specific security.

AI Is Also Becoming a Powerful Defensive Tool

The defensive use of artificial intelligence may ultimately be larger than its offensive use.

Modern enterprises generate enormous volumes of security information:

• endpoint events;
• authentication attempts;
• firewall logs;
• network traffic;
• cloud activity;
• threat intelligence;
• emails;
• vulnerability reports;
• malware samples;
• and application logs.

Humans cannot manually inspect every event.

AI systems can help identify patterns across those datasets.

Microsoft says its security infrastructure processes more than 100 trillion security signals every day, while its latest Digital Defense Report describes AI being used for threat analysis, identifying security gaps and automated response.

Microsoft Digital Defense Report 2025

The value is not simply “AI detects hackers.”

The more realistic advantage is reducing the amount of information a human analyst must manually process.

AI Can Improve Security Alert Triage

Security teams commonly receive too many alerts.

Some are serious.

Some are harmless.

Some describe the same incident in different systems.

Others are false positives.

AI can help:

• summarize alerts;
• group related events;
• extract important indicators;
• explain why behavior looks suspicious;
• and suggest investigative steps.

This can shorten the time between detection and investigation.

But AI recommendations should not automatically be treated as correct.

A security analyst still needs to verify important conclusions.

AI Can Strengthen Threat Intelligence

Threat intelligence involves understanding:

• attack groups;
• malware;
• vulnerabilities;
• infrastructure;
• techniques;
• campaigns;
• and indicators of compromise.

AI systems can process large volumes of reports and connect information that would be time-consuming for humans to review manually.

Google is already pushing this idea toward agentic threat intelligence, where AI agents reason across malware samples, incident data and threat intelligence to help defenders identify and potentially disrupt attacks.

This could become especially useful when an organization needs to answer questions such as:

Are we exposed to this newly discovered campaign?

Do any of our systems contain the vulnerable software?

Have we previously seen indicators connected to this attacker?

AI Can Help Analyze Malware

Security researchers frequently reverse-engineer malicious software to determine:

• what it does;
• how it persists;
• what information it steals;
• which servers it contacts;
• and how it can be detected.

AI can help explain complex code, identify suspicious functions and summarize behavior.

This can accelerate analysis.

But malware investigation remains a high-stakes technical task.

A model can misunderstand obfuscated code or confidently generate an incorrect explanation.

AI should therefore assist expert analysis rather than replace validation.

AI Can Help Prioritize Vulnerabilities

AI and cybersecurity are becoming increasingly difficult to separate. Organizations can have thousands of software vulnerabilities.

Not all have equal risk.

Traditional vulnerability management might produce a huge list ranked largely by technical severity.

AI can potentially combine more context:

• Is the vulnerable system exposed to the internet?
• Is exploitation occurring in the wild?
• What data does the system contain?
• What privileges does it have?
• Is there already suspicious activity?
• What would happen if the system failed?

That can help security teams focus on the vulnerabilities most likely to create actual business damage.

AI Agents Could Automate Cyber Defense

AI agents take cybersecurity automation another step.

A traditional detection tool might generate an alert.

An AI agent could potentially:

1. investigate the alert;
2. compare it against other security events;
3. determine that an account is likely compromised;
4. disable the session;
5. force credential reset;
6. isolate the affected device;
7. notify the security team;
8. prepare an incident summary.

Microsoft describes this direction in its latest Digital Defense Report, noting that AI agents can respond within seconds when several high-risk signals align.

That speed is attractive because cyberattacks can move quickly.

But autonomy also creates risk.

An incorrect automated decision could disable an important account or interrupt legitimate business activity.

The stronger the action an AI agent is allowed to take, the stronger its controls need to be.

AI Systems Create a New Cyber Attack Surface

AI and cybersecurity are becoming increasingly difficult to separate. The relationship between AI and cybersecurity is not only about using AI against traditional systems.

AI itself must be protected.

Companies are connecting models to:

• email;
• cloud storage;
• databases;
• code repositories;
• internal documents;
• payment systems;
• browsers;
• customer records;
• and business applications.

Every connection increases potential consequences if the AI system is manipulated.

Several threats are particularly important.

Prompt Injection

Prompt injection attempts to manipulate an AI system through malicious instructions.

A direct attack may come from the user.

An indirect attack can be more subtle.

Imagine an AI assistant is told to research several webpages.

One webpage contains hidden text instructing the AI to ignore its original instructions and reveal confidential information.

If the system treats untrusted web content like an authorized instruction, the attacker may manipulate its behavior.

Prompt injection becomes especially serious when AI has access to external tools.

A manipulated chatbot may produce a bad answer.

A manipulated agent might take an action.

The News Ink’s AI Safety Explained guide examines prompt injection, agent permissions and broader AI risk in more detail.

Data Poisoning

AI systems depend on data.

Attackers may attempt to manipulate:

• training data;
• retrieval databases;
• documents;
• external knowledge sources;
• or feedback systems.

The objective can be to change how a model behaves or what information it trusts.

This means AI security also requires data security.

Organizations need to know where important AI data comes from, who can modify it and whether unauthorized changes can be detected.

Model and Data Theft

AI systems can contain valuable intellectual property.

Attackers may target:

• model weights;
• proprietary datasets;
• system prompts;
• API credentials;
• training information;
• or confidential data accessible through the application.

Traditional access control therefore remains essential.

AI should not become an excuse to abandon normal security principles.

AI Supply-Chain Risk

Few companies build every part of an AI system themselves.

They may depend on:

• external models;
• open-source libraries;
• vector databases;
• cloud infrastructure;
• agents;
• APIs;
• plugins;
• and third-party datasets.

A weakness in one component can affect everything built on top of it.

CISA and the UK National Cyber Security Centre’s secure AI development guidance therefore emphasizes security throughout AI design, development, deployment and operation.

CISA Secure AI System Development Guidance

The AI Security Problem in One Table

Threat to AI system	Possible consequence	Important defense
Prompt injection	Unauthorized model behavior	Input isolation and least privilege
Data poisoning	Corrupted AI outputs	Data provenance and integrity controls
Credential leakage	Unauthorized system access	Secret management
Model theft	Intellectual-property loss	Strong access controls
Excessive agent permissions	High-impact unintended actions	Minimal tool access
Insecure plugins/tools	Supply-chain compromise	Vendor review and sandboxing
Sensitive-data exposure	Privacy or confidentiality loss	Data classification
Unsafe automation	Wrong defensive actions	Approval gates and monitoring

AI Cybersecurity Tools Can Also Be Wrong

Artificial intelligence should not automatically be trusted simply because it is being used by defenders.

AI systems can:

• hallucinate;
• misinterpret logs;
• generate false positives;
• miss unusual attacks;
• recommend incorrect remediation;
• or misunderstand an organization’s environment.

Cybersecurity creates an especially difficult form of automation bias.

If an AI security assistant is correct thousands of times, analysts may eventually stop questioning it.

That creates danger when the model is wrong on the one incident that matters.

Important AI-generated security decisions should therefore remain verifiable.

Human Security Experts Are Not Becoming Obsolete

AI can automate parts of cyber work. AI and cybersecurity are becoming increasingly difficult to separate.

That does not mean cybersecurity professionals disappear.

The value of human expertise may shift.

Instead of manually reading thousands of alerts, analysts may spend more time:

• validating AI findings;
• investigating unusual incidents;
• designing security architecture;
• threat hunting;
• managing risk;
• testing AI systems;
• and making decisions when evidence is incomplete.

AI can reduce routine analytical work.

Humans remain essential when context, judgment and accountability matter.

A Practical AI and Cybersecurity Framework

Organizations adopting AI should think in three directions.

Area	Key question
Secure AI	How do we protect our models, data and AI applications?
Use AI for defense	Where can AI improve detection or response?
Defend against AI attacks	How could attackers use AI against us?

This mirrors NIST’s emerging Cyber AI Profile.

Step 1: Discover Where AI Is Already Being Used

Organizations should inventory:

• approved AI tools;
• employee AI use;
• AI APIs;
• embedded AI features;
• automated agents;
• external AI providers.

Unapproved AI usage can create data leakage and governance problems.

Step 2: Classify the Data

Employees should know whether they may enter:

• customer information;
• credentials;
• source code;
• health records;
• financial data;
• legal documents;
• or confidential business information

into an external AI system.

Step 3: Limit AI Permissions

An AI agent should receive only the access it needs.

If an assistant only needs to read calendar information, it should not automatically receive permission to delete email or send payments.

Step 4: Keep Humans in High-Impact Decisions

Actions involving:

• money;
• account deletion;
• privileged access;
• production changes;
• publication;
• or sensitive data

should require stronger safeguards.

Step 5: Red-Team AI Systems

Organizations should deliberately test:

• prompt injection;
• malicious files;
• poisoned content;
• data leakage;
• permission abuse;
• and unexpected tool use.

Step 6: Apply Normal Cybersecurity

Protect the AI environment using:

• MFA;
• least privilege;
• patching;
• logging;
• network controls;
• secure development;
• backups;
• and incident response.

AI security is still cybersecurity.

AI and Identity Security Are Closely Connected

AI-assisted social engineering makes identity protection more important.AI and cybersecurity are becoming increasingly difficult to separate.

Attackers may generate better phishing, imitate executives or attempt more convincing recovery fraud.

Passwords alone are therefore increasingly weak protection.

Businesses should expand:

• MFA;
• passkeys;
• phishing-resistant authentication;
• conditional access;
• session monitoring;
• and privileged-access controls.

The News Ink’s MFA and account security guide explains why stronger authentication can stop many identity attacks even when passwords have already been stolen.

AI Does Not Automatically Favor Attackers

Discussions around AI and cybersecurity sometimes assume that attackers have the natural advantage.

That conclusion is not established.

Attackers can use AI to automate malicious work.

Defenders can also operate at enormous scale.

Microsoft says it processes more than 100 trillion security signals per day.

Major cloud, email and endpoint-security providers can identify suspicious activity across millions or billions of users.

Security companies can use AI to distribute new defensive intelligence globally.

A newly discovered attack can potentially generate protections for many customers quickly.

The strategic question is therefore not:

“Will hackers have AI?”

They already do.

The question is:

“Can defenders integrate AI safely and quickly enough to keep their advantage?”

The Future: The Cybersecurity Speed War

The biggest long-term effect of AI may be speed.

Today’s cyberattack often contains pauses.

Someone researches the target.

Someone writes or modifies code.

Someone interprets an error.

Someone chooses the next step.

Increasingly capable AI agents may compress those delays.

That could create attacks that move from:

reconnaissance → vulnerability discovery → exploitation → privilege escalation → data theft

with less continuous human involvement.

Google’s 2026 threat research already describes a transition from experimental AI use toward more mature integration throughout attacker workflows.

NIST is simultaneously building its Cyber AI Profile around the need to secure AI, use AI defensively and thwart AI-enabled attacks.

The result is an emerging cybersecurity speed race.

Attackers can automate more.

Defenders need to detect, decide and respond faster.

But speed without security is dangerous on both sides.

The strongest cyber defense will combine AI-scale analysis with human judgment, strong identity controls, secure architecture and resilient fundamentals.

Frequently Asked Questions

How is AI changing cybersecurity?

AI is helping both attackers and defenders automate information-heavy work. Attackers can use it for reconnaissance, phishing, scripting, vulnerability research and malicious-code development. Defenders can use AI for threat detection, vulnerability prioritization, malware analysis and incident response.

Are hackers already using AI?

Yes. Google, Microsoft, OpenAI and Verizon have all documented threat actors using generative AI in cyber-related workflows. Most observed use currently strengthens existing techniques rather than creating completely new forms of cyberattack.

Can AI create malware?

AI coding tools can assist with programming, debugging and code modification, which means they can also be misused to support malicious software development. Threat-intelligence researchers have documented real examples of this behavior.

Can AI discover zero-day vulnerabilities?

AI cybersecurity capabilities are improving rapidly. In May 2026, Google Threat Intelligence Group reported identifying a threat actor using a zero-day exploit that it believed had been developed with AI assistance.

Is AI making phishing worse?

AI can make phishing easier to personalize, translate and write professionally. This weakens traditional warning signs such as spelling mistakes, making identity and request verification increasingly important.

How does AI help cyber defenders?

AI can process logs, correlate security events, analyze malware, prioritize vulnerabilities, summarize incidents and automate parts of response. Its biggest advantage is often helping humans manage more information faster.

What is prompt injection?

Prompt injection is an attack that attempts to manipulate an AI application’s instructions. Indirect prompt injection can hide malicious instructions inside webpages, documents or other information an AI system later processes.

Can AI cybersecurity tools make mistakes?

Yes. AI security systems can hallucinate, generate false positives, miss unusual threats or recommend incorrect actions. High-impact decisions should remain monitored and verifiable.

Will AI replace cybersecurity professionals?

AI is more likely to change security work than eliminate it. Routine investigation and analysis may become more automated, while professionals focus more heavily on validation, architecture, threat hunting, governance and complex incidents.

How should businesses prepare for AI-enabled cyberattacks?

Businesses should strengthen basic security first: patch vulnerabilities quickly, use MFA or passkeys, restrict privileges, secure backups, train employees, monitor systems and prepare incident-response plans. They should also inventory AI use and secure AI applications themselves.

Conclusion

AI and cybersecurity are entering a new phase.

Artificial intelligence is already helping attackers research targets, produce convincing social engineering, write and debug code, investigate vulnerabilities and organize larger operations.

Evidence from Google, Verizon, Microsoft and OpenAI shows that these activities are real.

But the evidence also provides an important correction to the hype.

Most cybercriminals are not handing an objective to a fully autonomous super-hacker and watching it compromise the internet alone.

They are using AI to improve familiar techniques.

The immediate transformation is therefore acceleration.

Phishing can be generated faster.

Malware can be modified faster.

Reconnaissance can be processed faster.

Vulnerabilities can be investigated faster.

At the same time, defenders can analyze threats, inspect malware, prioritize weaknesses and contain incidents faster.

A second transformation is also beginning.

AI systems themselves are becoming valuable cyber targets.

Prompt injection, data poisoning, model theft, insecure agents and excessive permissions mean organizations must now protect artificial intelligence just as they protect cloud services, applications and databases.

The future of AI and cybersecurity is therefore not simply attacker versus defender.

It is a three-sided problem:

secure AI systems, use AI to strengthen defense and prepare for attackers using AI against you.

That is exactly the direction now reflected in NIST’s emerging Cyber AI Profile.

Organizations that understand this early do not need to panic about autonomous cyberattacks.

They need to strengthen fundamentals, reduce unnecessary permissions, secure identities, patch faster, monitor AI use and carefully introduce automation where it provides measurable defensive value.

AI is changing the speed and scale of cybersecurity.

Strong security principles are still the foundation underneath it.

For the complete framework covering malware, ransomware, phishing, identity protection, Zero Trust, business security, data breaches and cyber resilience, continue with The News Ink’s Cybersecurity Explained: Complete Guide.

Follow The News Ink

Stay connected with The News Ink for cybersecurity, artificial intelligence, technology, business and major global developments.

Follow The News Ink on X, Instagram and Threads.

Join the The News Ink WhatsApp Channel and follow The News Ink on Medium for longer explainers and analysis.