Ref. Ares(2020)2922680 - 05/06/2020
Twitter also announced in April that, between July and December 2017, 274,460
accounts were suspended for violations related to promotion of terrorism, and of
those suspensions 93% consisted of accounts flagged by internal, proprietary
spam-fighting tools, while 74% of those accounts were suspended before their
first tweet.
There is stil more to be done in engaging smaler platforms, but we have made
a good start.
In addition, the military campaign against Daesh has brought down the level of
production of official propaganda from its peak.
However, a smaler volume of content, much of which is recycled, or unofficial
supporter created content, is now dispersed across a greater number of platforms.
Our analysis shows that between July and December 2017 Daesh used 145
platforms which were not used in the previous six months; and used approximately
400 unique platforms in 2017. Over 100 platforms have been used by Daesh so far
in 2018 to disseminate and host propaganda online.
3. Do you see a risk that removal by companies on their own initiative could interfere
with investigations or intelligence gathering? What would be the mitigating
measures necessary to address any such risks?
4. Do you see a risk of erroneous removal by platforms of legal content (e.g. removal of
content misidentified as ilegal, removal of content disseminated for research,
educational or journalistic purposes, "over-removal")? Are you aware of any cases of
over-removal? What would be the mitigating measures necessary to address any such
risks?
5
There is a smal, manageable risk. However, automation technology can be trained to
make a distinction between legal journalistic or counter-narrative content and terrorist
propaganda. The Home Office developed technology with ASI Data Science that has
been adversarially trained against educational and journalistic content to enable it to
make this distinction. Tests have shown this new tool can automatically detect 94%
of Daesh propaganda with 99.995% accuracy. This means an error rate of just 1 in
20,000.
Despite this, a process for redress is required to ensure that, if content is mistakenly
identified as ilegal content, it is not permanently removed from the platform.
6
referral mechanism, avoiding unnecessary duplication of efforts, facilitating
requests from law enforcement agencies in relation to criminal investigations4.
X
Stronger commitment on specific proactive and preventive measures (i.e. further
development and participation in industry- led schemes, such as the database of
hashes developed in the context of the EUIF)
X
More detailed requirements on transparency and reporting
X
More detailed requirements to companies on safeguards against over-removal
X
Establishment of an external audit/monitoring mechanism
X
Establishment of contact points, both in companies and Member States, to
facilitate referrals (and feedback) and requests from law enforcement authorities
in relation to criminal investigations.
X
Additional support (e.g. by Europol) to referral capacities in Member States
4. What other additional measures could be developed within a reinforced voluntary
approach?
Key measures set out above.
5. Which further actions could be taken to secure participation from those companies who
have not engaged?
6. Which further actions could be taken to support small companies and start-ups in
tackling terrorist content online effectively? Should these be taken by larger companies,
public authorities or both?
We believe the most effective way is through industry-led engagement by larger
CSPs. This is because they have the expertise to help, and also because small
companies can be reluctant to engage with public authorities.
7. Do you think that the voluntary approach is effective and flexible enough to ensure that
companies continue their efforts in the long term? Please indicate with which statement
you would agree with:
4 See point 40 of the Recommendation.
8
Yes
No, it should be reinforced as presented above to obtain sufficient
guarantees
X No, it should be reinforced via legislation
III.
Legislative options
1. Why would you consider legislation necessary at this time? What would be the
concrete benefits? What risks could legislation entail?
The UK Government announced on 20 May that it wil bring forward online safety
legislation that would cover the ful range of online harms, including online terrorist
content.
Any action taken must:
have a meaningful impact on the problem we are trying to solve, and
must be more effective than the current voluntary approaches. We
want companies to focus on:
transparency to ensure that effective measures are being taken by
the companies to remove ilegal content and to show compliance;
and
speed of removals, as our research shows that approximately a
third of al links to Daesh propaganda are disseminated within an
hour of release;
take into consideration smaler platforms which are increasingly
being used to host and share online terrorist content as the larger
companies have improved their response;
be multi-jurisdictional; and
be future proofed.
2. What should be the material scope of legislation (i.e. how should terrorist content be
defined)? Do you consider that covering material inciting to commit terrorist acts
(Article 21/Article 5 of the Terrorism Directive5) is sufficient or should the
dissemination of material pursuing other terrorist purposes be included as wel ?
5 Directive (EU) 2017/541 of the European Parliament and of the Council of 15 March 2017 on combating
terrorism and replacing Council Framework Decision 2002/475/JHA and amending Council Decision
2005/671/JHA
https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32017L0541
9
Material the dissemination of which pursues the following objectives should be
included in legislative measures:
X Recruitment for terrorism
X Providing training for terrorism
X Terrorist financing
Other, please elaborate:
To what extent should material produced by UN/EU designated terrorist organisatio ns
be included?
Fuly. In addition, the legislation should make provision to cover new and quickly
evolving terrorist groups and their material. EU and UN lists are a valuable basis of
reference, but we should be ready to react to material produced by new groups that
are not yet officially prohibited. Consideration should also be given to groups
proscribed in individual member states, such as National Action in the UK.
3. Which measures (based in particular on the elements mentioned in the Inception Impact
Assessment) do you consider as necessary elements of legislation to be impactful?
Please indicate the need from a scale from 1 (unnecessary) to 5 (very necessary)
10
Do you consider that minimum requirements could usefully be complemented by self-
regulatory measures? And if so, which ones?
4. What other additional measures could be developed within legislation?
5. What should be the personal scope of the legislation? Only hosting service providers
within the meaning of the Directive on electronic commerce or other service providers?
6 See point 41 of the Recommendation.
7 See points 16 and 17 of the Recommendation.
11
6. Do you think smaller companies should be covered by al obligations or should they
be exempted from some of the obligations (e.g. proactive measures) but obliged by
others (e.g. time-limits after referral)? Which companies could be partially exempted
and from which obligations?
7. How do you see the impact on fundamental rights of the above-mentioned measures
and which safeguards would be necessary to avoid undue interference with fundame nta l
rights?
Article 6 of the Charter of Fundamental Rights of the European Union states that
everyone has the right to liberty and security of person. Taking action against il ega l
content online aims to enhance security.
We are of the view that the impact on fundamental rights is negligible.
12