Digitisation in Agriculture

The traditional way of doing agriculture has undergone massive digitization in recent years, whereby several agricultural processes have been linked to the Internet. This globally prevalent transformation, driven by smart technology, encompasses the use of sensors, IoT devices, and data analytics to optimize and automate labour-intensive farming practices. Smart farmers in the country and abroad now leverage real-time data to monitor soil conditions, weather patterns, and crop health, enabling precise resource management and improved yields. The integration of smart technology in agriculture not only enhances productivity but also promotes sustainable practices by reducing waste and conserving resources. As a result, the agricultural sector is becoming more efficient, resilient, and capable of meeting the growing global demand for food.

Digitisation of Food Supply Chains

There has also been an increase in the digitisation of food supply chains across the globe since it enables both suppliers and consumers to keep track of the stage of food processing from farm to table and ensures the authenticity of the food product. The latest generation of agricultural robots is being tested to minimise human intervention. It is thought that AI-run processes can mitigate labour shortage, improve warehousing and storage and make transportation more efficient by running continuous evaluations and adjusting the conditions real-time while increasing yield. The company Muddy Machines is currently trialling an autonomous asparagus-harvesting robot called Sprout that not only addresses labour shortages but also selectively harvests green asparagus, which traditionally requires careful picking. However, Chris Chavasse, co-founder of Muddy Machines, highlights that hackers and malicious actors could potentially hack into the robot's servers and prevent it from operating by driving it into a ditch or a hedge, thereby impending core crop activities like seeding and harvesting. Hacking agricultural pieces of machinery also implies damaging a farmer’s produce and in turn profitability for the season.

Case Study: Muddy Machines and Cybersecurity Risks

A cyber attack on digitised agricultural processes has a cascading impact on online food supply chains. Risks are non-exhaustive and spill over to poor protection of cargo in transit, increased manufacturing of counterfeit products, manipulation of data, poor warehousing facilities and product-specific fraud, amongst others. Additional impacts on suppliers are also seen, whereby suppliers have supplied the food products but fail to receive their payments. These cyber-threats may include malware(primarily ransomware) that accounts for 38% of attacks, Internet of Things (IoT) attacks that comprise 29%, Distributed Denial of Service (DDoS) attacks, SQL Injections, phishing attacks etc. 

Prominent Cyber Attacks and Their Impacts

Ransomware attacks are the most popular form of cyber threats to food supply chains and may include malicious contaminations, deliberate damage and destruction of tangible assets (like infrastructure) or intangible assets (like reputation and brand). In 2017, NotPetya malware disrupted the world’s largest logistics giant Maersk and destroyed all end-user devices in more than 60 countries. Interestingly, NotPetya was also linked to the malfunction of freezers connected to control systems. The attack led to these control systems being compromised, resulting in freezer failures and potential spoilage of food, highlighting the vulnerability of industrial control systems to cyber threats.

Further Case Studies

NotPetya also impacted Mondelez, the maker of Oreos but disrupting its email systems, file access and logistics for weeks. Mondelez’s insurance claim was also denied since NotPetya malware was described as a “war-like” action, falling outside the purview of the insurance coverage. In April 2021, over the Easter weekend, Bakker Logistiek, a logistics company based in the Netherlands that offers air-conditioned warehousing and food transportation for Dutch supermarkets, experienced a ransomware attack. This incident disrupted their supply chain for several days, resulting in empty shelves at Albert Heijn supermarkets, particularly for products such as packed and grated cheese. Despite the severity of the attack, the company successfully restored their operations within a week by utilizing backups. JBS, one of the world’s biggest meat processing companies, also had to pay $11 million in ransom via Bitcoin to resolve a cyber attack in the same year, whereby computer networks at JBS were hacked, temporarily shutting down their operations and endangering consumer data. The disruption threatened food supplies and risked higher food prices for consumers. Additional cascading impacts also include low food security and hindrances in processing payments at retail stores. 

Credible Threat Agents and Their Targets

Any cyber-attack is usually carried out by credible threat agents that can be classified as either internal or external threat agents. Internal threat agents may include contractors, visitors to business sites, former/current employees, and individuals who work for suppliers. External threat agents may include activists, cyber-criminals, terror cells etc. These threat agents target large organisations owing to their larger ransom-paying capacity, but may also target small companies due to their vulnerability and low experience, especially when such companies are migrating from analogous methods to digitised processes.

The Federal Bureau of Investigation warns that the food and agricultural systems are most vulnerable to cyber-security threats during critical planting and harvesting seasons. It noted an increase in cyber-attacks against six agricultural co-operatives in 2021, with ancillary core functions such as food supply and distribution being impacted. Resultantly, cyber-attacks may lead to a mass shortage of food not only meant for human consumption but also for animals.

Policy Recommendations

To safeguard against digital food supply chains, Food defence emerges as one of the top countermeasures to prevent and mitigate the effects of intentional incidents and threats to the food chain. While earlier, food defence vulnerability assessments focused on product adulteration and food fraud, including vulnerability assessments of agriculture technology now be more relevant.  

Food supply organisations must prioritise regular backups of data using air-gapped and password-protected offline copies, and ensure critical data copies are not modifiable or deletable from the main system.  For this, blockchain-based food supply chain solutions may be deployed, which are not only resilient to hacking, but also allow suppliers and even consumers to track produce. Companies like Ripe.io, Walmart Global Tech, Nestle and Wholechain deploy blockchain for food supply management since it provides overall process transparency, improves trust issues in the transactions, enables traceable and tamper-resistant records and allows accessibility and visibility of data provenance. Extensive recovery plans with multiple copies of essential data and servers in secure, physically separated locations, such as hard drives, storage devices, cloud or distributed ledgers should be adopted in addition to deploying operations plans for critical functions in case of system outages. For core processes which are not labour-intensive,  including manual operation methods may be used to reduce digital dependence. Network segmentation, updates or patches for operating systems, software, and firmware are additional steps which can be taken to secure smart agricultural technologies. 

References

1. Muddy Machines website, Accessed 26 July 2024.  https://www.muddymachines.com/ 
2. “Meat giant JBS pays $11m in ransom to resolve cyber-attack”, BBC, 10 June 2021. https://www.bbc.com/news/business-57423008 
3. Marshall, Claire & Prior, Malcolm, “Cyber security: Global food supply chain at risk from malicious hackers.”, BBC, 20 May 2022. https://www.bbc.com/news/science-environment-61336659 
4. “Ransomware Attacks on Agricultural Cooperatives Potentially Timed to Critical Seasons.”, Private Industry Notification, Federal Bureau of Investigation, 20 April  https://www.ic3.gov/Media/News/2022/220420-2.pdf. 
5. Manning, Louise & Kowalska, Aleksandra. (2023). “The threat of ransomware in the food supply chain: a challenge for food defence”, Trends in Organized Crime. https://doi.org/10.1007/s12117-023-09516-y 
6. “NotPetya: the cyberattack that shook the world”, Economic Times, 5 March 2022. https://economictimes.indiatimes.com/tech/newsletters/ettech-unwrapped/notpetya-the-cyberattack-that-shook-the-world/articleshow/89997076.cms?from=mdr 
7. Abrams, Lawrence, “Dutch supermarkets run out of cheese after ransomware attack.”, Bleeping Computer, 12 April 2021. https://www.bleepingcomputer.com/news/security/dutch-supermarkets-run-out-of-cheese-after-ransomware-attack/ 
8. Pandey, Shipra; Gunasekaran, Angappa; Kumar Singh, Rajesh & Kaushik, Anjali, “Cyber security risks in globalised supply chains: conceptual framework”, Journal of Global Operations and Strategic Sourcing, January 2020. https://www.researchgate.net/profile/Shipra-Pandey/publication/338668641_Cyber_security_risks_in_globalized_supply_chains_conceptual_framework/links/5e2678ae92851c89c9b5ac66/Cyber-security-risks-in-globalized-supply-chains-conceptual-framework.pdf 
9. Daley, Sam, “Blockchain for Food: 10 examples to know”, Builin, 22 March 2023 https://builtin.com/blockchain/food-safety-supply-chain 

‍

Introduction
‍

Agentic AI systems are autonomous systems that can plan, make decisions, and take actions by interacting with external tools and environments. But they shift the nature of risk by blurring the lines among input, decision, and execution. A conventional model generates an output and stops. An agent takes input, makes plans, invokes tools, updates its state and repeats the cycle. This creates a system where decisions are continuously revised through interaction with external tools and environments, rather than being fixed at the point of input.

This means the attack surface expands in size and becomes more dynamic. Instead of remaining confined to components as in traditional computational systems, they spread in layers and can continue to grow through time. To understand this shift, the system can be analysed through functional layers such as inputs, memory, reasoning, and execution, while recognising that risk does not remain isolated within these layers but emerges through their interaction.

‍

‍

Agentic AI Attack Surface

‍

A layered view of how risks emerge across input, memory, reasoning, execution, and system integration, including feedback loops and cross-system dependencies that amplify vulnerabilities.
‍

Input Layer: Where Untrusted Data Becomes Control
‍

The entry point of an agent is no longer one prompt. The documents, APIs, files, system logs and the outputs of other agents can now be considered input. This diversity is significant due to the fact that every source of input carries its own trust assumptions, and in the majority of cases, they are weak.

The most obvious threat is prompt injection, where inputs are treated as instructions rather than data. Since inputs are treated as instructions, a virus, a malicious webpage, or a document can contain instructions that override system goals without necessarily being detected as something harmful.

Indirect prompt injection extends this risk beyond direct user interaction. Instead of targeting the interface, attackers compromise the retrieval process by embedding malicious instructions within external data sources. When the agent retrieves and processes the data, it treats the embedded content as legitimate input. As a result, the attack is executed through normal reasoning processes, allowing the system to act on untrusted data without recognising the manipulation.

Data poisoning also occurs at runtime. In contrast to classical poisoning (where training data is manipulated), runtime poisoning distorts the agent’s perception of its environment as it runs. This can change decisions without causing apparent failures.

Obfuscation introduces another indirect attacker vector. Encoded instructions or complicated forms may bypass human review but remain readable to the model. This creates asymmetry whereby the system knows more about the attack than those operating it. Once compromised at this layer, the agent implements compromised instructions which affect downstream operations.
‍

Context and Memory: Persistence of Influence
‍

Agentic systems depend on memory to operate efficiently. They often retain context across sessions and frequently store information between sessions.

This introduces a different type of risk: persistence. Through memory poisoning, attackers can insert false or adversarial information into sorted context, which then influences future decisions. Unlike prompt injection, which is often limited to a single interaction, this effect carries forward. Over time, the agent begins to operate on a distorted internal state, shaping decisions in ways that may not be immediately visible.

Another issue is cross-session leakage. Information in a particular context may be replayed in a different context when memory is being shared or there is insufficient memory separation. This is specifically dangerous in those systems that combine retrieval and long-term storage. The context management in itself becomes a weakness. Agents are required to make decisions on what to retain and what to discard. This is susceptible to attackers who can flood the context or manipulate what is still visible and indirectly affect reasoning.

The underlying problem is structural. Memory turns data into a state. Once state is corrupted, the system cannot easily distinguish valid knowledge from adversarial influence.

The issue is structural. Memory converts temporary data into a persistent state. Once this state is weakened, the system cannot reliably separate valid information from adversarial influence, making recovery significantly more difficult.
‍

Reasoning and Planning: Manipulating Intent Without Breaking Logic
‍

The reasoning layer is where agentic AI stands apart from traditional systems. The model no longer reacts to inputs alone. It actively breaks down objectives, analyses alternatives, and ranks actions.

At the reasoning stage, the nature of risk shifts. The concern is no longer limited to injecting instructions, but to influencing how decisions are made. One example is goal manipulation, where the agent subtly reinterprets its objective and produces outcomes that are technically correct but strategically harmful. Reasoning hijacking operates within intermediate steps, altering how constraints are evaluated or how trade-offs are prioritised. The system may remain internally consistent, which makes such deviations difficult to detect.

Tool selection becomes a critical control point. Agents decide which tools to use and when, so influencing these choices can redirect execution without directly accessing the tools themselves. Hallucinations also take on a different role here. In static systems, they remain errors. In agentic systems, they can trigger actions. A perceived need or incorrect judgement can translate into real-world consequences.

This layer introduces probabilistic failure. The system is not fully weakened, but it is nudged towards decisions that appear reasonable yet are incorrect. The risk lies in how those decisions are justified.

‍
‍Tool and Execution: When Decisions Gain Reach
‍

Once an agent begins interacting with tools, its behaviour extends beyond the model into external systems. APIs, databases, and services become part of the execution path.

One key risk is the use of unauthorised tools. When agents operate with broad permissions, any manipulation of the upstream can be converted into real-world actions. This makes access control a central security concern. Command injection also takes a different form here. The agent generates commands based on its reasoning, so if that reasoning is compromised, the resulting actions may still appear valid despite being harmful.

External tool outputs introduce another risk. If these systems return corrupted or misleading data, the agent may accept it without verification and incorporate it into its decisions. It is also becoming increasingly reliant on third-part tools and plugins adds to this exposure. If these components are compromised, they can affect behaviour without directly attacking the core system, creating a supply-side risk.

At this stage, the agent effectively operates as an insider. It holds legitimate credentials and interacts with systems in expected ways, making misuse harder to identify.
‍

Application and Integration: System-Level Exposure
‍

Agentic systems rarely operate in isolation. They are embedded in larger environments, interacting with identity systems, business logic, and operational workflows.

Access control becomes a major vulnerability. Agents tend to operate across multiple systems with various permission models, creating irregularities that can be exploited. Risks also arise from identity and delegation. In case an agent is operating on behalf of a user, then any vulnerabilities in authentication or session management can allow attackers to assume that authority.

Workflow execution amplifies these risks. Agents can initiate multi-step processes such as transactions, updates, or approvals. Manipulating a single step can change the result of the entire workflow. As integrations increase, so do the number of interaction points, making cumulative risk harder to track.

At this layer, failures are not isolated. They propagate into business operations, making consequences harder to contain.
‍

Output and Action: Where Failures Become Visible
‍

The output layer is where failures become visible, though they rarely originate there.

Data leakage has been a key concern. Agents may disclose information they are allowed to access, especially when tasks boundaries are not clearly defined. Misinformation and unsafe outputs are also important, particularly when outputs directly influence actions or decisions.

Generated code and commands introduce execution risk. If outputs are used without validation, errors or manipulations can have system-level effects. The shift towards autonomous action increases this risk, as small upstream deviations can lead to significant consequences without human intervention. This layer reflects symptoms rather than root causes. Addressing it alone does not reduce the underlying risk.
‍

Beyond Layers: The Missing Dimension
‍

A layered view helps, but it does not capture the full picture. Agentic systems are defined by continuous interaction across layers.

The key missing dimension is the runtime loop. Inputs shape reasoning, reasoning drives action, and actions feed back into both reasoning and memory. These cycles create feedback loops, where small manipulations may escalate over time. This also reduces observability. With multiple interacting components, it becomes difficult to trace cause and effect or identify where failures originate.

Supply chain dependencies add another layer of risk. Models, datasets, APIs, and plugins each introduce their own points of failure. A compromise at any of these points can propagate across the system. The attack surface also includes governance. Weak supervision, unclear responsibility, or excessive autonomy increase overall risk. Human control is not external to the system; it is part of its security.
‍

Conclusion: Structuring the Attack Surface
‍

Agentic AI expands the attack surface beyond traditional systems. It is both recursive and stateful. Risk does not just accumulate across layers; it moves and changes as the system operates.

Any useful representation must go beyond a linear stack. It should capture feedback loops, persistent state, and cross-layer dependencies that characterise the way these systems actually behave. The system is not a pipeline but a cycle. That is where both its capability and its risk emerge.

‍

Introduction and Brief Analysis

A movie named “The Artifice Girl” portrayed A law enforcement agency developing an AI-based personification of a 12-year-old girl who appears to be exactly like a real person. Believing her to be an actual girl, perpetrators of child sexual exploitation were caught attempting to seek sexual favours. The movie showed how AI aided law enforcement, but the reality is that the emergence of Artificial Intelligence has posed numerous challenges in multiple directions. This example illustrates both the promise and the complexity of using AI in sensitive areas like law enforcement, where technological innovation must be carefully balanced with ethical and legal considerations.

Detection and Protection tools are constantly competing with technologies that generate content, automate grooming and challenge legal boundaries. Such technological advancements have provided enough ground for the proliferation of Child Sexual Exploitation and Abuse Material (CSEAM). Also known as child pornography under Section 2 (da) of Protection of Children from Sexual Offences Act, 2012, it defined it as - “means any visual depiction of sexually explicit conduct involving a child which includes a photograph, video, digital or computer-generated image indistinguishable from an actual child and image created, adapted, or modified, but appears to depict a child.”

Artificial Intelligence is a category of technologies that attempt to shape human thoughts and behaviours using input algorithms and datasets. Two Primary applications can be considered in the context of CSEAM: classifiers and content generators. Classifiers are programs that learn from large data sets, which may be labelled or unlabelled and further classify what is restricted or illegal. Whereas generative AI is also trained on large datasets, it uses that knowledge to create new things. Majority of current AI research related to AI for CSEAM is done by the use of Artificial neural networks (ANNs), a type of AI that can be trained to identify unusual connections between items (classification) and to generate unique combinations of items (e.g., elements of a picture) based on the training data used.

Current Legal Landscape

‍

The legal Landscape in terms of AI is yet unclear and evolving, with different nations trying to track the evolution of AI and develop laws. However, some laws directly address CSEAM. The International Centre for Missing and Exploited Children (ICMEC) combats Illegal sexual content involving children. They have a “Model Legislation” for setting recommended sanctions/sentencing. According to research performed in 2018, Illegal sexual content involving children is illegal in 118 of the 196 Interpol member states. This figure represents countries that have sufficient legislation in place to meet 4 or 5 of the 5 criteria defined by the ICMEC.

CSEAM in India can be reported on various portals like the ‘National Cyber Crime Reporting Portal’. Online crimes related to children, including CSEAM, can be reported to this portal by visiting cybercrime.gov.in. This portal allows anonymous reporting, automatic FIR registration and tracking of your complaint. ‘I4C Sahyog Portal’ is another platform managed by the Indian Cyber Crime Coordination Centre (I4C). This portal integrates with social media platforms.

The Indian legal front for AI is evolving and CSEAM is well addressed in Indian laws and through judicial pronouncements. The Supreme Court judgement on Alliance and Anr v S Harish and ors is a landmark in this regard. The following principles were highlighted in this judgment.

‍

• The term “child pornography” should be substituted by “Child Sexual Exploitation and Abuse Material” (CSEAM) and shall not be used for any further judicial proceeding, order, or judgment. Also, parliament should amend the same in POCSO and instead, the term CSEAM should be endorsed.
• Parliament to consider amending Section 15 (1) of POCSO to make it more convenient for the general public to report by way of an online portal.
• Implementing sex education programs to give young people a clear understanding of consent and the consequences of exploitation. To help prevent Problematic sexual behaviour (PSB), schools should teach students about consent, healthy relationships and appropriate behaviour.
• Support services to the victims and rehabilitation programs for the offenders are essential.
• Early identification of at-risk individuals and implementation of intervention strategies for youth.

‍

Distinctive Challenges

According to a report by the National Centre for Missing and Exploited Children (NCMEC), a significant number of reports about child sexual exploitation and abuse material (CSEAM) are linked to perpetrators based outside the country. This highlights major challenges related to jurisdiction and anonymity in addressing such crimes. Since the issue concerns children and considering the cross-border nature of the internet and the emergence of AI, Nations across the globe need to come together to solve this matter. Delays in the extradition procedure and irregular legal processes across the jurisdictions hinder the apprehension of offenders and the delivery of justice to victims.

CyberPeace Recommendations

For effective regulation of AI-generated CSEAM, laws are required to be strengthened for AI developers and trainers to prevent misuse of their tools. AI should be designed with its ethical considerations, ensuring respect for privacy, consent and child rights. There can be a self-regulation mechanism for AI models to recognise and restrict red flags related to CSEAM and indicate grooming or potential abuse.

A distinct Indian CSEAM reporting portal is urgently needed, as cybercrimes are increasing throughout the nation. Depending on the integrated portal may lead to ignorance of AI-based CSEAM cases. This would result in faster response and focused tracking. Since AI-generated content is detectable. The portal should also include an automated AI-content detection system linked directly to law enforcement for swift action.

Furthermore, International cooperation is of utmost importance to win the battle of AI-enabled challenges and to fill the jurisdictional gaps. A united global effort is required. Using a common technology and unified international laws is essential to tackle AI-driven child sexual exploitation across borders and protect children everywhere. CSEAM is an extremely serious issue. Children are among the most vulnerable to such harmful content. This threat must be addressed without delay, through stronger policies, dedicated reporting mechanisms and swift action to protect children from exploitation.

‍

References:

• https://www.sciencedirect.com/science/article/pii/S2950193824000433?ref=pdf_download&fr=RR-2&rr=94efffff09e95975
• https://aasc.assam.gov.in/sites/default/files/swf_utility_folder/departments/aasc_webcomindia_org_oi d_4/portlet/level_2/pocso_act.pdf
• https://www.manupatracademy.com/assets/pdf/legalpost/just-rights-for-children-alliance-and-anr-vs-s‍harish-and-ors.pdfhttps://www.icmec.org‍https://www.missingkids.org/theissues/generative-ai

‍