=Paper=
{{Paper
|id=Vol-2844/ethics9
|storemode=property
|title=Artificial Intelligence and Whistleblowing: Can A.I. be useful for Whistleblowing processes? (short paper)
|pdfUrl=https://ceur-ws.org/Vol-2844/ethics9.pdf
|volume=Vol-2844
|authors=Kalliopi Zouvia
|dblpUrl=https://dblp.org/rec/conf/setn/Zouvia20
}}
==Artificial Intelligence and Whistleblowing: Can A.I. be useful for Whistleblowing processes? (short paper)==
https://ceur-ws.org/Vol-2844/ethics9.pdf
Artificial Intelligence and Whistleblowing: Can A.I. be useful
for Whistleblowing processes?
Kalliopi Zouvia
Lawyer, DPO and Senior Investigator
Greek Ombudsman
PhD Candidate Panteion University
of Social and Political Sciences
zouvia@synigoros.gr
laws and regulations existed in the various states, more action was
ABSTRACT needed at national level in order to implement them effectively
In recent years, there is a constantly growing discussion about the [5]. In 2010 the Assembly of the Council of Europe adopted
contribution of whistleblowers and the need to protect them. Resolution 1729, calling on all Member States to review their
Another rising topic of discussion is the increasing use of whistleblower protection legislation following specific guidelines
Artificial Intelligence (AI), the gains from it, the hardships, the [6].
moral dilemmas. This article refers to the development of
In 2014 the Council of Europe developed a legal instrument on
whistleblower protection in the EU and discusses how AI can be
protecting individuals who report or disclose information on acts
applied in a whistleblowing context.
and omissions in the workplace that represent a serious threat or
harm to the public interest. The Recommendation 2014/7 set out a
CCS Concepts series of principles to guide member States when reviewing their
• Applied Computing ➝ Law, social and behavioral national laws or when introducing legislation and regulations or
sciences ➝ Law making amendments that are necessary and appropriate in the
context of their legal system [7].
• Computing Methodologies ➝ Artificial Intelligence
It was after major scandals, such as LuxLeaks, that the European
Parliament took steps towards the protection of whistleblowers
and in 2017 the plenary Assembly adopted a Resolution on
Keywords legitimate measures to protect whistleblowers acting in the public
Artificial Intelligence; Whistleblowing; Law Technology. interest when disclosing confidential information of companies
and public bodies [8].
1. INTRODUCTION Finally, in April 2019, the EU passed the Whistleblower Directive
In the last decade a lot of important whistleblower cases came to or Directive 2019/1937 of the European Parliament and of the
light, such as the E. Snowden case [1], Lux Leaks [2], Panama Council on the protection of persons who report breaches of
Papers, the Novartis case. In the last few months the value of Union Law and outlines a basic standard of protection that will
whistleblowing became even more obvious in the context of apply across all 27 member states of the EU [9]. The Directive sets
battling the global COVID-19 pandemic. Dr Li Wenliang was one rules for who can be protected as a whistleblower and the kinds
of a group of medics who were reprimanded by local Chinese of issues they can make reports on. It defines how the idea of
authorities for sharing information about the early cluster of protecting whistleblowers interacts with other legal and ethical
patients with SARS-like symptoms admitted to Wuhan hospitals obligations. The Directive needs to be transposed into national
in December 2019. A few months later, healthcare professionals, legislation by the end of December 2021.
in several countries around the world, came forward with alerts
about shortages of personal protective equipment, which were The Directive refers to protecting individuals who report breaches
forcing them to work in unsafe conditions [3]. of EU Law and its purpose, as set out in Article 1, is to enhance
the ‘enforcement of Union law and policies in specific areas’, such
One of the definitions of the term whistleblowing is “the as public procurement, financial services, safety of products,
disclosure by organization members (former or current) of illegal, transport safety, protection of the environment.
immoral or illegitimate practices under the control of their
employers, to persons of organizations that may be able to effect A lot has already been written in the aftermath of the adoption of
action” [4]. the Directive [10]. In this article only a few key points are going
to be mentioned, in order then to examine whether artificial
The phenomenon of individuals coming forward to speak about intelligence could be used in whistleblower protection.
important issues has a long history, but the term ‘whistleblowing’
came to prominence in the US in the 1960 and early 70’s. Some of Regarding the reporting channels, the Directive follows the three
the fist law intended to protect whistleblowers emerged in the US. –tiered model [11], which means that whistleblowers can report
internally [12], within a legal entity, externally to competent
In Europe, the Committee on Legal Affairs and Human Rights of authorities [13] or, as a last resort, they can disclose their
the Parliamentary Assembly of the Council of Europe submitted, information publicly, to the media [14].
in 2009, a report which concluded that although whistleblower
WAIEL2020, September 3, 2020, Athens, Greece
Copyright © 2020 for this paper by its authors. Use permitted under Creative
Commons License Attribution 4.0 International (CC BY 4.0).
Confidentiality and anonymity are really important for the “Digital Single Market” [19]. AI can be used to bring the
whistleblower protection. When we talk about confidentiality it benefits of the technology to society and economy, for citizens,
means that the identity of a whistleblower is known to specific for business development and for services of public interest [20].
individuals, whereas anonymity means that a whistleblower is not As far as whistleblowing is concerned, thoughts are already being
known at all. According to the Directive ‘Member States shall made about the use of Artificial Intelligence in whistleblowing
ensure that the identity of the reporting person is not disclosed to practice [21]. In the following sections, we elaborate on
anyone beyond the authorized staff members competent to possibilities of using AI for whistleblowing in three areas:
receive or follow up on reports, without the explicit consent of reporting systems, vetting process automation and
that person. This shall also apply to any other information from proactive monitoring.
which the identity of the reporting person may be directly or
indirectly deduced’ [15]. As far as anonymity is concerned, the 2.1 AI for whistleblowing reporting systems
Directive lets Member States decide whether organizations and Advances in technology are already influencing changes in the
competent authorities are required to accept and follow up on way speak up reports are performed. AI can be a technology
anonymous reports. This is not in line with current best practice. enabler for whistleblower reporting systems in several ways.
While anonymous disclosures can make it harder to investigate a Virtual AI agents (chatbots) can assist or entirely handle the
concern, this should not prevent a concern being taken seriously submission of whistleblower reports. Chatbots are AI-enabled
[10]. In order to secure the identity of the whistleblower there are software applications that can conduct conversations with human
hotlines and technology has been a game changer in anonymous users with voice or text [22]. They use natural language
whistleblowing, with the advent of the anonymous online drop processing (NLP) capabilities to recognize what their human
boxes, which use encryption and other privacy enhancing interlocutors say and to formulate responses that try to resemble
technologies to obscure the identity of the reporting person. what a human agent would reply. Chatbots are widely used today
A timely response after a report is also important. According to in a range of industries, for example for customer care in
the Directive, a reporting person is entitled to have their report telecommunications, banking, and healthcare. They are
acknowledged within seven days and receive feedback within particularly effective as interactive guides of users to structured
three months [12]. processes or transactions, such as submitting an issue report to a
There is not much data about whistleblower reports in Europe. telecommunications provider, reporting a lost credit card or
One of the few countries for which there is data is the scheduling a medical appointment. Virtual agents can be helpful
Netherlands: according to the Dutch Whistleblowers Authority for whistleblowers submitting an allegation by interactively
2018 Annual Report, a total of fifty requests for investigation had providing instructions throughout the process. They can also help
been submitted between 2016 and 2018. Of these, 28 were declared reduce the number of incomplete or non-eligible reports by
inadmissible [16]. advising users on what are the requirements for a submission and
the scope of handled cases. A recent study reports that
In the US, however, there is an increase in whistleblower reports whistleblowers were more likely to report to an online platform
in the last few years. According to the 2019 US Securities and when a virtual agent handled the reports because they believed
Exchange Commission Annual Report to Congress on the that it is more efficient and provides greater control while
Whistleblower Program, which provides incentives and reporting [23].
protection to whistleblowers, the Commission received more than
5.000 reports, which represents a 74% increase since AI-based live translation can help increase the accessibility of
whistleblower data collection started in 2012 [17]. whistleblowing reporting services across ethnic and cultural
communities. Automatic real time translation can enable
The Occupational Safety and Health Administration, a interactive cross-language reporting through hotlines, i.e. over
government agency responsible for protecting workers in a wide phone to a human agent or through a chatbot (voice or text).
range of industries, received in 2018 more than 9.500 reports, Voice-based hotlines are a very common submission channel for
creating a significant backlog in light of a reduction in the number whistleblowers. For example, it is by far the most frequent
of investigators. More than 3.000 of those complaints resulted in a reporting method (37% of reports in 2019) in corporate ethics
full investigation. With such great workload, it is important to compliance systems, as reported by the largest provider of ethics
have the means to react quickly and in the most appropriate way. reporting software in the US [24]. Real-time machine translation
This raises the question whether Artificial Intelligence could be of is a relatively mature technology, fueled particularly by advances
use. in neural networks and deep learning [25]. There are commercial
2. ARTIFICIAL INTELLIGENCE FOR offerings of translation-as-a-service by major technology
providers which can be used by the developers of whistleblowing
WHISTLEBLOWER PROTECTION applications to allow cross-lingual communication between
submitters of whistleblowing reports and handing agents on a
Artificial Intelligence is distinguished in two big general 24/7 basis without the huge costs of having stand-by human
categories. Artificial Narrow Intelligence (ANI), sometimes interpreters for multiple languages [23]. There are already real-
referred to as Weak AI, that specializes in one area and life applications of live translation in other domains, such as
Artificial General Intelligence (AGI), sometimes referred to as communication of teachers with parents of different cultures in
Strong AI, or Human-Level AI. Artificial General Intelligence middle schools in the US [26].
refers to a computer that is as smart as a human across the board
— a machine that can perform any intellectual task that a human 2.2 AI for making the vetting process more
being can [18]. efficient
AI is a collection of technologies that combines data, algorithms As seen from the above-mentioned scandals and a lot of others
and computing power. The evolution of Artificial Intelligence has that see the light all the more often the last years, there is a lot of
been rapid in recent years with the result that it is at the center of misconduct. While whistleblower protection evolves, many
organizations will engage in some type of whistleblowing
investigation, whether it is conducted internally or from an 3. CONCLUSIONS – FUTURE OUTLOOK
outside agency. It is important for an organization to give the tone
from the top and cultivate a culture that promotes ethical behavior
and a speak-up culture. Being able to understand the motives of More organized steps were taken just recently in order to better
whistleblowers is also important, but vetting and investigating the regulate whistleblower protection in the EU and already there are
complaints is critical. To this end, one must have the appropriate thoughts about the use of AI in whistleblower protection. Even
auditing tools. the idea that robots could replace whistleblowers is being
explored as a potential future development [31]. If robots
Currently, the impact of AI is mostly discussed in financial audits
gradually take the place of workers (let’s say in the automotive
and is especially pronounced in the area of data acquisition (data
industry), will there be a time that robots conceive and report
extraction, comparison, and validation). This means that AI-
wrongdoings? One could argue that the ‘penalisation’ of the
enabled technology can locate relevant information, extract it
whistleblower will no longer exist [32].
from documents, and make it usable for the human auditor, who
can devote more time to areas requiring higher-level judgment. Nevertheless, the potential replacement of whistleblowers by AI
For example, AI enables full automation of time-consuming tasks raises a lot of questions, some of them being whether a robot will
such as payment transaction testing, including extraction of any know what to blow the whistle on and if it will be able to follow
supporting data for further substantive testing [27][28]. the three tiered model. Moreover, depending on the country and
legislation, whistleblowers can report illegalities, irregularities,
The procedural aspect of a whistleblower investigation is more or
wrongdoings or immoral actions. Can a robot know immoral
less the same as any other investigation that corporations are
actions? [31] Scandals as LuxLeaks did not entail illegal acts but
subjected to. What is critical in an internal whistleblower
obscure legal practices that were considered immoral by society.
investigation is to determine if the allegation is true [29].
Will a machine ever be able to report such acts even internally?
Whistleblower reports are usually followed by a great amount of
Artificial Intelligence is a useful tool in detecting wrongdoing.
data. Sometimes though only a small part of this data may actually
Nevertheless, a fair balance between human actions and the use
be relevant to the action or omission that represents a serious
of AI will always be a crucial aspect.
threat or harm. Advanced analytic tools and AI can be used in
order to locate and extract critical information from
whistleblower reports. Artificial Intelligence in the form of
machine learning can group similar documents for faster review. 4. REFERENCES
Data sampling is useful to provide insight into larger data.
One could distinguish between the use of AI for structured data [1] Hackett, K. (2013). Edward Snowden: The New Brand of Whistle-
analytics and its use for analytics over unstructured data [30]. Blower? Quill Magazine. 101, 5 (Sept/Oct 2013), 26-31.
Structured data is comprised of clearly defined data types whose [2] Tansini, C. (2018). The Luxleaks Whistleblowing Case: a Philosophical
pattern makes them easily searchable; while unstructured data – Assessment.
“everything else” – is comprised of data that is usually not as [3] Abazi, V. (2020). Truth distancing? Whistleblowing as remedy to censorship
easily searchable, including formats like audio, video, and social during covid-19. European Journal of Risk Regulation, 11(2), 375–381.
media postings
[4] Near, J.P., Miceli, M.P. (1985) Organizational dissidence: The case of whistle-
In structured data analytics, AI and analytics tools can be set up blowing. Journal of Business Ethics 4, 1–16.
to look for example for transactions that exceed norms, [5] Protection of “whistle-blowers”, Doc. 12006 (14 September 2009), Report of
transactions with vague or missing detail, rush requests, unusual the Committee on Legal Affairs and Human Rights,
cash disbursements, or transactions that circumvent typical [http://assembly.coe.int/nw/xml/XRef/Xref-XML2HTML-en.asp?fileid=12302]
approval processes. All these could be characterized as red flags [6] Vandekerckhove, W. and Lewis, D. 2011. The Content of Whistleblowing
and could direct the investigator towards the person or persons Procedures: A Critical Review of Recent Official Guidelines. Journal of
responsible for those transactions. On the other hand, in Business Ethics. 108, 2 (Nov. 2011), 253–264.
unstructured data (documents, email, and other messaging) [7] Protection of whistleblowers, Recommendation CM/Rec (2014) 7 adopted by
keyword language filters could be used in order to expose use or the Committee of Ministers of the Council of Europe on 30 April 2014 and
change in language that may indicate unethical or noncompliant explanatory memorandum [http://rm.coe.int/doc/09000016807096c7]
behavior. [8] European Parliament resolution of 24 October 2017 on legitimate measures to
protect whistle-blowers acting in the public interest when disclosing the
2.3 AI for proactive monitoring confidential information of companies and public bodies (2016/2224(INI))
Artificial Intelligence can also be used for proactive monitoring, https://www.europarl.europa.eu/doceo/document/TA-8-2017-0402_EN.html
in order to identify potential areas of risk. If for example a certain [9] Directive 2019/1937 of the European Parliament and of the Council of 23
type of bad behavior is becoming more common in an October 2019 on the protection of persons who report breaches of Union law
organization’s industry sector, that particular subject matter could [2019] OJ L 305/17 https://eur-lex.europa.eu/legal-
content/en/TXT/?uri=CELEX%3A32019L1937
be targeted for routine monitoring using AI and linguistic and
analytics tools to flag worrisome language or sentiment in data [10] Terracol, M. 2019. Building on the EU Directive for Whistleblower Protection,
Analysis and Recommendation. Transparency International Position Paper #1
stores. Analytical tools and predictive software will enable / 2019. ISBN: 978-3-96076-127-3.
companies to combine whistleblowing data with information
from across the business and identify where problems are most [11] Vandekerckhove, W. 2010. European Whistleblower Protection: Tiers or
Tears? in D. Lewis (ed.), A Global Approach to Public Interest
likely to occur in the future [28]. Disclosure (Cheltenham: Edward Elgar, 2010).
[12] Chapter II Internal Reporting and Follow Up, Art. 7-9 in Directive 2019/1937
of the European Parliament.
[13] Chapter III External Reporting and Follow Up, Art. 10-14 in Directive
2019/1937 of the European Parliament.
[14] Chapter IV, Public Disclosures, Art. 15 in Directive 2019/1937 of the [24] Penman, C. (2020). 2020 Risk & Compliance Hotline Benchmark Report.
European Parliament. Navex Global Benchmark Reports. Retrieved from
https://www.navexglobal.com/en-gb/resources/benchmarking-reports/2020-
[15] Chapter V, Art. 16 Duty of confidentiality in Directive 2019/1937 of the risk-compliance-hotline-benchmark-report
European Parliament.
[25] Koehn, P. (2020). Neural machine translation. Cambridge University Press.
[16] Dutch Whistleblowers Authority. March 2019. Annual Report 2018. ISBN: 9781108601764.
https://www.huisvoorklokkenluiders.nl/Publicaties/jaarverslagen/2019/03/14/
annual-report-2018---dutch-whistleblowers-authority [26] Le Santo, G. Real-time language translation boosted by AI. Orange Business
Services Blogs. Online article: https://www.orange-
[17] US Securities and Exchange Commission. 2019. Whistleblower Program 2019 business.com/en/blogs/language-translation-boosted-by-ai
Annual Report to the US Congress. https://www.sec.gov/files/sec-2019-
annual-report-whistleblower-program.pdf [27] Kokina, J. and Davenport, T.H. 2017. The Emergence of Artificial Intelligence:
How Automation is Changing Auditing. Journal of Emerging Technologies in
[18] Tassis S. 2018. The era of Artificial Intelligence, Media and Communication Accounting. 14, 1 (Mar. 2017), 115–122.
Law (ΔιΜΕΕ) 4/2018, 484-494.
[28] Ghanoum, S., & Alaba, F. M. 2020. Integration of Artificial Intelligence in
[19] Voras A., Mitrou L. 2018. Artificial Intelligence and Data Protection: A view Auditing : The Effect on Auditing Process. 1–75. Master’s Thesis. Kristianstad
under the light of the General Data Protection Regulation. Media and University. Spring 2020.
Communication Law (ΔιΜΕΕ) 4/2018, 460-466.
[29] Grobart, J. 2020. Preparing for Whistleblower Complaints. Strategic Finance.
[20] European Commission. European Data Strategy. COM(2020) 66 final. September 2020. https://sfmagazine.com/post-entry/september-2020-
https://eur-lex.europa.eu/legal- preparing-for-whistleblower-complaints/
content/EN/TXT/?uri=CELEX%3A52020DC0066
[30] O’Leary, D.E. 2013. Artificial Intelligence and Big Data. IEEE Intelligent
[21] Mohd Noor, N.R.A. and Mansor, N. 2019. Exploring the Adaptation of Systems. 28, 2 (Mar. 2013), 96–99. DOI:https://doi.org/10.1109/mis.2013.39.
Artificial Intelligence in Whistleblowing Practice of the Internal Auditors in
Malaysia. Procedia Computer Science. 163, (2019), 434–439. [31] Kafteranis, D. 2019. Can artificial intelligence replace whistle-blowers in the
business sector. International Journal of Technology Policy and Law. 3, 2
[22] Dale, R. 2016. The return of the chatbots. Natural Language Engineering 22(5) (2019), 160. DOI:https://doi.org/10.1504/ijtpl.2019.104948.
811-817.
[32] Waytz, A. 2014. Why Robots Could Be Awesome Whistleblowers. The
[23] Gao, L. 2020. Technology Advancements for Whistleblowing Reporting Atlantic. October 2014.
Platforms and Employees’ Decision to Blow the Whistle. Accounting and the https://www.theatlantic.com/business/archive/2014/10/why-robots-could-be-
Public Interest. (Oct. 2020). DOI:https://doi.org/10.2308/api-2020-001 awesome-whistleblowers/381216/