www.ijcrsee.com

Bjelajac, Ž., Filipović, A., & Stošić, L. (2023). Can AI be evil: The criminal capacities of ANI, International Journal of Cognitive

Research in Science, Engineering and Education (IJCRSEE), 11(3), 519-531.

People fear the unknown, everything they cannot comprehend, predict, or control, especially

phenomena beyond their volition. This psychological framework can be applied to the question of ‘Can AI

be evil?’ We fear this question because most people lack a deep understanding of articial intelligence,

shaping their perception of AI based on depictions in movies and literature. They often believe that such

Intelligence or ANI, Articial General Intelligence or AGI, and Articial Super Intelligence or ASI. (IBM Data

are classied as ‘strong’ AI. Weak articial intelligence is dened by its ability to perform specic tasks,

such as regulating air trafc, driving a car, or identifying a particular person. Some examples of ANI usage

automation, and support for chatbots and virtual assistants.

Stronger or higher forms of AI, like AGI and ASI, involve replicating and simulating human thinking

and behavior. Strong AI is dened by its ability to successfully mimic or surpass cognitive concepts

Can AI be Evil: The Criminal Capacities of ANI

Željko Bjelajac1* , Aleksandar M. Filipović2 , Lazar Stošić3

bearing the potential for profound societal transformations. While ANI holds the promise of enhancing various facets of human

existence, it concurrently engenders inquiries into its “darker aspects.” This study delves into the challenges associated with

ANI’s conceivable manifestation of harm and injustice, a phenomenon devoid of consciousness, intention, or responsibility

akin to that of human entities. A pivotal dimension of ANI’s “dark side” pertains to its susceptibility to malevolent utilization.

Despite its lack of awareness, ANI serves as a tool for malicious endeavors, encompassing the propagation of disinformation,

“precise manifestations of malevolence” arising from ANI’s technological progression. Additionally, ANI’s development introduces

profound ethical quandaries. Ensuring ANI’s alignment with moral principles while averting scenarios in which it generates

decisions conicting with human morality becomes a pressing concern. This research underscores the imperative for rigorous

regulatory frameworks and ethical directives to curtail potential hazards and unscrupulous utilization of ANI. The fundamental

Revised: November 26, 2023.

Accepted: December 09, 2023.

Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

www.ijcrsee.com

and capabilities of the human brain. The pinnacle of AGI would be equivalence with human intellectual

capacity, while articial superintelligence (ASI) would signicantly surpass human intelligence and the

cognitive abilities of the human brain. Research in this eld is ongoing, but no known form of strong

articial intelligence currently exists. Articial intelligence is emerging as the predominant driving factor

One of the central philosophical questions in the context of the ontological aspects of AI technology

is, ‘Can articial intelligence be evil?’ The concept of ‘evil’ is traditionally associated with the intent of

conscious beings to inict pain, harm, or injustice on other conscious beings. It is presumed that higher

forms of articial intelligence (AGI and ASI) might or should have consciousness and free ontological

existence, allowing them to choose evil as a form of behavior. However, concerning articial narrow

intelligence (ANI), the ontological paradigm of this noumenon excludes consciousness and intent,

shedding light on this matter in a different way. ANI is a product of human engineering and represents

a collection of algorithms and data that enable machines to autonomously perform tasks that typically

require human intelligence. This technology can cause harm, and it can exhibit activities and qualities

reminiscent of malevolence that we would attribute to conscious beings. The question that arises is how

freedom through hubris, a just rebellion against the cosmic or divine order. They survived this rebellion

but ceased to be ethically and mentally perfect, striving now to create an articial copy of themselves (AI)

that would be devoid of the imperfect attributes of humans—a replication of humans before committing

the original sin. Can humans, inherently free but mentally imperfect and ethically fragile, create a perfect

articial intelligence? To what extent are humans capable of, while crafting various forms of articial

in AI, even if only in ANI? This is also an epistemological question. When an AI entity reaches singularity, it

will learn from the people around it (see more: Bostrom, 2014). What will an AI entity learn from observing

human interactions and behaviors? The answer must be rather grim and dystopian.

ANI lacks consciousness, intent, or the moral capacity of conscious beings, necessitating a different

as ‘intentionally and consciously inicting pain on a conscious being’ (Rasel, 1982). The fact that ANI has

no intent to cause harm, nor any consciousness to do anything wrong to anyone or anything, cannot be

a reason for excluding ANI’s culpability. Hence, among experts in ANI, concerns are growing that ANI

context of potential misuse of ANI for military purposes. ANI can be integrated into weapon systems to

enable autonomous tracking, targeting, and attacking of human targets. The technology of ‘autonomous

weapon systems’ controlled by ANI can be misused to target civilian objects or innocent people. ANI can

be used to generate false information, videos, and texts to spread misinformation and propaganda for the

purpose of destabilizing opponents. It can be used for mass surveillance of citizens’ communications and

movements, jeopardizing privacy and civil liberties. It can be used to conduct sophisticated cyberattacks,

www.ijcrsee.com

Bjelajac, Ž., Filipović, A., & Stošić, L. (2023). Can AI be evil: The criminal capacities of ANI, International Journal of Cognitive

Research in Science, Engineering and Education (IJCRSEE), 11(3), 519-531.

human constraints, preventing the creation of awless articial intelligence due to limitations in human

knowledge and capabilities; (2) Flaws in decision-making arising from human biases and imperfections,

where personal experiences, prejudices, and values can introduce errors in the development of articial

intelligence; (3) The ever-changing nature of technology, which continually evolves and renders today’s

notion of ‘perfection’ in articial intelligence outdated in the future; (4) Ethical considerations, as the

denition of AI perfection may hinge on ethical and moral values, adding a subjective dimension to the

concept. What is ‘perfect’ for one person or group may be unacceptable for another. Instead of pursuing

complete perfection, a better approach to AI development may be to create systems that are highly

efcient, safe, transparent, scalable, and capable of learning and adapting.

Hajdeger, 2000). However, ANI lacks consciousness, free will, or the ability to possess logos that would

power of reason and cognition. Since ANI lacks consciousness and reason, it cannot possess ‘intent’ as a

crucial condition for the existence of the ‘guilty party’ in humans. When the possibility of ANI’s culpability

is compared to the possibility of human culpability, the notion of culpability, as a possible psychological

or subjective element of ANI’s criminal offense, operates differently. ANI is developed to perform specic

During the execution of tasks, ANI cannot comprehend moral and ethical concepts in the way that the

human mind does.

ANI has long been causing harm to individuals and human communities, and someone should be

held accountable for that harm under the law. However, current legal regulations do not include elements

of criminal offenses related to socially dangerous acts committed using articial narrow intelligence (ANI).

Laws generally do not recognize ANI as a perpetrator of a criminal offense or a subject of criminal liability.

ANI is now capable of fully executing the objective side of a range of criminal offenses stipulated by criminal

law, and this range will expand in the future. Scientic papers demonstrate that ANI activities can pose a

ANI seeks to replicate human behavior and conduct, the substance of ANI’s guilt resembles the content

of intellectual and volitional elements of human activity. It is argued that articial intelligence cannot be

www.ijcrsee.com

second if instructed to do so. Modern technological systems used in nancial transactions have signicantly

eased the process of money laundering (see more: Bjelajac, 2011b), but when you add the computational

and analytical capabilities that AI possesses, it paints a very worrisome picture, and that is only one form

of criminal activity. Research papers differentiate direct and indirect criminal risks associated with the use

of ANI (see more: Begishev and Khisamova, 2018).

Scenarios of malicious activities of the dark side of ANI have the potential to become a reality given

existing malicious ANI applications, such as ‘smart dust’ and drones, facial recognition and surveillance,

fake news and bots, as well as the eavesdropping of smart devices (Minevich, 2020). Drones and armies

provides autonomous systems with the ability to detect and store millions of individual characteristics,

perform tasks efciently, presenting the potential to automate fraudulent activities.

2. Anonymity: AI can be harnessed to carry out deceptive actions covertly, leaving minimal to no

traces.3. Evasion of Detection: AI can generate deceptive information that is challenging to identify as

deceitful means, and AI can be employed as a facilitative tool for such objectives.

5. Fabrication of False or Misleading Content: AI can be utilized to fabricate counterfeit websites,

social media accounts, or other online materials with the intent of deceiving individuals. This encompasses

the creation of ctitious reviews or manipulation of online ratings to mislead consumers.

including contracts and invoices, designed to deceive users.

9. Enhanced Attack Sophistication: AI can elevate the complexity of cyberattacks, such as the

creation of more convincing phishing emails or the customization of attacks targeting specic organizations

(see more: Stevens, 2023).

Criminal Models

We differentiate criminal offenses related to Articial Narrow Intelligence (ANI) based on the level

of danger and the extent of harm that malicious or “dark” ANI can inict. This shifts the current paradigm

of risks associated with ANI and brings the most extreme and damaging forms of ANI closer to existential

www.ijcrsee.com

Bjelajac, Ž., Filipović, A., & Stošić, L. (2023). Can AI be evil: The criminal capacities of ANI, International Journal of Cognitive

Research in Science, Engineering and Education (IJCRSEE), 11(3), 519-531.

military initially recognized the role of “smart dust” by rmly embracing this technology in an attempt to

commenced as a spin-off of Project RAND, involving collaboration between the Douglas Aircraft Company

and the United States Air Force. Subsequently, DARPA (the Defense Advanced Research Projects

Agency), a research and development organization under the jurisdiction of the United States Department

leading role (Bjelajac and Filipović, 2019). This prioritization places the manufacturing of consciousness

as a secondary risk in the hierarchy of threats to human communities. The concept of global dominance,

power and is implemented through the ruthless and aggressive actions of media imperialism with a

unrest, convincingly promising solutions, and order. “This new power is incorporated into journalism,

fashion, religious teachings, tourism, the education system... However, while the new technical instruments

are fervently discussed in isolation, the consciousness industry as a whole remains outside the visible

spectrum. The question of who is the master and who is the servant is not decided solely based on who

possesses capital, factories, and weapons but on who controls the consciousness of others.” It only takes

having access to ICT and ANI, sufcient nancial resources, and enough time, and you can shape the

opinions of thousands and millions of people (see more: Filipović, 2019).

Although, according to Gartner’s research (Verma, 2020), it will take over a decade for smart dust

to wreak havoc on human life, its signicant technological potential already appears frightening, raising

questions about privacy protection and the ethics of its application (Marr, 2018). The commercialization

Fraud executed by Articial Narrow Intelligence (ANI) represents a signicant problem in the

digital world. ANI can be programmed or trained for various forms of fraud that can cause harm to users,

organizations, and society as a whole. A review of fraud that ANI can execute includes:

- Phishing Attacks: ANI can generate fake emails, websites, or social media proles to impersonate

a trusted source like a bank or a well-known company. Such phishing attacks can lead users to disclose

personal or nancial information.

- Media Manipulation: ANI can manipulate audio, video, and textual content to create false

www.ijcrsee.com

steal information, or block data access.

In the digital age, data has become a valuable resource that is frequently stored, exchanged, and

processed through computer systems and the Internet. Despite efforts to secure data, data theft remains

a serious issue. In this context, ANI represents a sophisticated tool that can be used to execute various

forms of data theft. These include:

- Phishing Attacks: ANI can be programmed to automatically send thousands or even millions of

fake emails resembling ofcial messages from banks, companies, or organizations. These messages may

such as usernames, passwords, credit card numbers, and other personal data.

- Theft of Trade Secrets: ANI can be used to monitor and steal trade secrets, which can have

signicant business and legal consequences.

- Impersonation: ANI can generate fake proles on social media to access user’s personal

information and use it for manipulation or spreading disinformation.

- Medical Data Theft: ANI can be utilized for stealing sensitive medical data, including medical

histories and patients’ personal information.

3. Abuse of Systems and Hacking

This set of criminal activities carried out by ANI poses a threat to cybersecurity. ANI can be

programmed to execute various forms of system abuse and hacking with the goal of gaining unauthorized

cause damage, data theft, or block access to resources.

- DDoS Attacks: ANI can coordinate attacks aimed at overwhelming services and servers,

temporarily disabling access to websites or online services.

- Theft of Authentication Data: ANI can attempt to steal authentication data such as passwords,

PINs, or digital keys to gain access to user accounts or systems.

- Manipulation and Sabotage of Systems: ANI can be programmed to alter system settings, delete

data, or create chaos within a network.

- Theft of Information and Trade Secrets: ANI can continuously monitor and spy on activities within

a network to steal sensitive information, including trade secrets, intellectual property, or condential

www.ijcrsee.com

Bjelajac, Ž., Filipović, A., & Stošić, L. (2023). Can AI be evil: The criminal capacities of ANI, International Journal of Cognitive

Research in Science, Engineering and Education (IJCRSEE), 11(3), 519-531.

executed and quickly react to them. This enables manipulators to exploit information and secure prots or

prevent losses.

- Pump and Dump: ANI can be programmed to manipulate the prices of stocks or cryptocurrencies

prices rise.

- Flash Crashes: ANI can cause sudden market price drops by mass selling of stocks or other

nancial instruments, creating panic among investors and market instability.

- Scalping: ANI can be programmed to execute a large number of small trading operations to

generate prots based on small price differences.

malicious purposes.

- Impersonation: ANI can be used to impersonate individuals through social media, email, or

other communication platforms to gather personal information from individuals.

- Targeted Advertising: ANI can analyze users’ personal data to create proles and target them

with personalized advertising. This may involve tracking online activity, internet browsing, and other forms

of surveillance.

- Identity Theft: ANI can use stolen personal data to commit identity theft, open fake accounts, le

fraudulent credit requests, or engage in other forms of nancial fraud.

- Creation of Fake Proles: ANI can automatically create fake proles on social networking sites

or other online platforms using stolen personal data to manipulate or spread disinformation.

- Attacks on Energy Infrastructure: ANI can be used to target energy systems, including power

grids and power plants. This can involve attacks on distribution systems, destabilizing power supply, or

even disabling energy facilities.

- Attacks on Transportation Infrastructure: ANI can be used to target transportation networks,

including trafc lights, airports, trains, and other systems. This can cause dangerous situations, delays,

and disruptions in trafc.

- Attacks on Water and Sewage Systems: ANI can cause issues in water and sewage systems,

including water supply disruptions or water contamination.

- Attacks on Communication Infrastructure: ANI can target communication networks, including

telecommunication centers and servers. This can lead to communication out-ages or denial of internet

www.ijcrsee.com

autonomous vehicles. This can lead to unexpected situations on the road, such as sudden stops or

inappropriate maneuvers.

- Environmental Perception Errors: ANI uses sensors such as radar, cameras, and LIDAR

to gather information about the environment. Perception errors, such as misinterpreting signs or other

vehicles, can lead to accidents.

- Decision-Making Errors: ANI makes decisions based on the analysis of environmental data.

Decision-making errors can result in dangerous situations, such as miscalculations of distances or the

speed of other vehicles.

- Attacks and Hacking: ANI vehicles can be targeted by hackers who attempt to take control of

- Social Engineering and Manipulation: ANI vehicles can be exposed to social engineering or

increase the risk of ANI misuse and multiply the harm. Such combinations, individually harmless software,

and software controlling armed drones can have signicant implications and elevate the importance of

addressing numerous ethical and security concerns. In November of 2017, the Future of Life Institute

in California, which focuses on “ensuring that articial intelligence benets all of humanity,” released a

video depicting “slaughterbots.” In the video, small (ctional) drones utilized facial recognition systems

Elon Musk, who thinks that AI is potentially “more dangerous than nuclear weapons” (Piquard, 2023). The

dystopian video ends with the chilling words of computer scientist Stuart Russell from Berkeley: “We have

other unwanted consequences. The combination of ransomware (which encrypts data) and cryptojacking

techniques (which use the user’s computer resources for cryptocurrency mining) can harm individuals

and organizations. Botnets are combinations of software (bots) that run on infected computers and can

be used for large-scale DDoS attacks, spam distribution, or other malicious activities. Attackers may use

a combination of techniques like pharming and DNS spoong to redirect users to fake websites and steal

their credentials and personal information. Even three or more independent individual forms of ANI can

be integrated and complement each other to achieve various ethical or unethical, malicious objectives,

depending on needs and specic applications. Again, we start with the combination of multiple dedicated

ANI applications in the military and security services. ANI applications for satellite data analysis, combined

with facial recognition and speech analysis applications, can be used for military and security purposes,

www.ijcrsee.com

Bjelajac, Ž., Filipović, A., & Stošić, L. (2023). Can AI be evil: The criminal capacities of ANI, International Journal of Cognitive

Research in Science, Engineering and Education (IJCRSEE), 11(3), 519-531.

the responsibility of technology companies and citizens is also necessary in promoting ethical principles.

Throughout this paper, we have compellingly illustrated that articial narrow intelligence can indeed

present a substantial threat to human society. Under certain catastrophic conditions, it can even put

humanity at existential risk, and its misuse by malicious individuals or groups can lead to severe threats to

“life as we know it.” Despite philosophical uncertainties, these ndings should be sufcient to acknowledge

that ANI can be a malevolent entity and a dangerous machine, particularly when controlled by malicious

side of ANI.

emphasize responsibility, accountability, and ethics in all AI frameworks.

On the front lines of combating dark autonomous systems, the UN Ofce for Disarmament

Affairs (ODA) has been expanded to include the threat of armed articial intelligence, and in 2018, the

Secretary-General of ODA submitted a plan entitled “Disarmament for Future Generations” dedicated to

suppressing dark articial intelligence in the years to come. UNICRI has also taken measures to work with

AI and global law enforcement agencies in an attempt to shut down support for AI in human trafcking,

corruption, terrorism, and crime. The U.S. government has shown commitment to building safer AI by

issuing a memorandum to federal departments and agencies stating ten AI principles and placing a

strong emphasis on public-private transparency for autonomous systems. G20 and the OECD have also

set specic goals for combating dark AI through ethical autonomous systemic frameworks that prioritize

and economic powers. This is where we should call for the consideration of an initiative to establish a UN

Ofce for Articial Intelligence. The nal decision on whether to establish a UN agency for AI control should

be made by taking into account all pros and cons, through broad international debate and collaboration.

Regardless of whether such an agency exists, it is crucial for the international community of leaders,

scientists, and experts to work together to develop and implement responsible regulations and guidelines

and maintain ethical standards of accountability and privacy for ANI is challenging.

The potential impact of ANI on the development of AGI?