Democracy Reporting International Sri Lanka

12/08/2026

⏳ Deadline extended!

There’s still time to apply for the External Project Evaluator Consultancy under our project, Strengthening Resilience Against Disinformation in Sri Lanka.

We’re looking for an individual or team to conduct a comprehensive review of our project’s delivery, assessing its relevance, effectiveness, methodological rigour, and overall quality through document review and consultations with our partners, beneficiaries, and other relevant stakeholders.

🗓️ New application deadline: 19 August 2026, 18:00 CET

Apply before the extended deadline: https://tinyurl.com/37zz3bv9

Know someone who’d be a great fit? Share this opportunity with your network.

11/08/2026

A national election may involve millions of voters. But online, the conversation was driven by a remarkably small number of voices.

Our latest analysis found that 10 influencer accounts produced 81.3% of all election-related content, while the top three alone accounted for 47.4%. The findings reveal just how concentrated political discourse became across digital platforms among a handful of influential accounts.

Discover what this means for digital campaigning and electoral discourse in our latest “Breaking the Silence” Report. Link here: https://tinyurl.com/yz7bfh4p

🤝 Co-funded by the European Commission and implemented together with our partner Factum

ජාතික මැතිවරණයකට ඡන්දදායකයන් මිලියන ගණනක් සම්බන්ධ විය හැකිය. එහෙත් අන්තර්ජාලය තුළ, එම සංවාදය මෙහෙයවන්නේ ඉතා සුළු හඬවල් කිහිපයක් විසිනි.

අපගේ නවතම විශ්ලේෂණයෙන් හෙළි වූයේ, මැතිවරණයට අදාළ සමස්ත facebook සහ YouTube අන්තර්ගතයෙන් 81.3%ක්ම නිර්මාණය කර කර ඇත්තේ influencer ගිණුම් 10ක් විසින් වන අතර, ඉන් ඉහළම ගිණුම් තුනක් 47.4%ක අන්තර්ගතයන් නිර්මාණය කර තිබෙනවා. ඩිජිටල් වේදිකා හරහා දේශපාලන කතිකාවත, අතලොස්සක් වූ influencer ගිණුම් කිහිපයක් අතර කෙතරම් දුරට සංකේන්ද්‍රණය වී ඇත්ද යන්න මෙම වාර්තාවේ සොයාගැනීම් මඟින් අනාවරණය වේ.

ඩිජිටල් මැතිවරණ ව්‍යාපාර සහ මැතිවරණ කතිකාව සඳහා මෙහි ඇති වැදගත්කම කුමක්දැයි අපගේ නවතම "Breaking the Silence" වාර්තාවෙන් සොයා ගන්න. සබැඳිය මෙන්න: https://tinyurl.com/yz7bfh4p

🤝 යුරෝපා සංගමයේ සම-මුල්‍ය ආධාර ඇතිව සහ අපගේ හවුල්කරු වන Factum සමඟ ඒකාබද්ධව ක්‍රියාත්මක කරයි

Tamil

தேர்தல் ஒன்றில் இலட்சக்கணக்கான வாக்காளர்கள் பங்குபெறக்கூடும். எனினும் இணையத்தில், அத்தகைய தேர்தல் தொடர்பான உரையாடலை மிகச் சில குரல்களே முன்னெடுத்தன.

எங்களது புதிய பகுப்பாய்வின்படி, தேர்தல் தொடர்பான மொத்த Facebook மற்றும் YouTube உள்ளடக்கங்களில், 10 influencer கணக்குகளினால் மாத்திரம் 81.3% உள்ளடக்கங்கள் உருவாக்கப்பட்டுள்ளதுடன், அவற்றில் முதல் மூன்று கணக்குகள் மாத்திரம் 47.4% உள்ளடக்கங்களை உருவாக்கியுள்ளன. டிஜிட்டல் தளங்கள் ஊடாக அரசியல் விவாதங்கள், ஒரு சில influencer கணக்குகளுக்குள் மாத்திரம் எவ்வளவு தூரம் குவிந்துள்ளன என்பதை இப்புதிய அறிக்கையின் கண்டுபிடிப்புகள் வெளிப்படுத்துகின்றன.

டிஜிட்டல் தேர்தல் பிரச்சாரங்களிலும், தேர்தல் விவாதங்களிலும் இதன் தாக்கம் என்ன என்பதை எமது புதிய அறிக்கை "Breaking the Silence" மூலம் கண்டறியுங்கள். அறிக்கையை இவ்விணைப்பினூடாக காணலாம்: https://tinyurl.com/yz7bfh4p

இத்திட்டம் ஐரோப்பிய ஒன்றியத்தின் நிதிஉதவியுடன் எமது பங்காளரான Factum உடன் இணைந்து செயற்படுத்தப்படுகிறது.

07/08/2026

⏳ Time is running out.

Applications for the External Project Evaluator for our project are closing soon.

We're looking for an individual/ team to conduct a comprehensive review of the project's delivery, assessing its relevance, effectiveness, methodological rigour, and overall quality through document review and consultations with our partners and beneficiaries, including the Election Commission of Sri Lanka.

🗓️ Applications close: 12 August 2026, 18:00 CET

Apply before the deadline: https://tinyurl.com/37zz3bv9

Know someone who'd be a great fit? Share this opportunity with your network.

05/08/2026

Not every voice shaping Sri Lanka's election conversation belongs to a political party, or a newsroom.

Our new study examines who these voices actually are. We reviewed 4,219 pieces of Facebook and YouTube content from Sri Lanka's 2024 presidential election, examining posting volume, platform distribution, and how candidates were discussed.

The findings point to several clear patterns.

Facebook accounted for more than 93% of all monitored content, confirming its position as the primary platform for election-related discussion. Ten influencers were responsible for 81.3% of all election-related content, indicating a high degree of concentration among a small number of accounts. Election-related content remained prominent even during the legally mandated silence period. Two accounts alone accounted for close to a third of all election posts published during that period.

Candidates Anura Kumara Dissanayake, Ranil Wickremesinghe, Sajith Premadasa and Namal Rajapaksa were mentioned consistently throughout the monitoring period, through a combination of promotion, criticism and commentary.

Together, these findings indicate that a relatively small group of influencers holds considerable sway over which candidates and messages reach voters online.

Full study is available here. https://tinyurl.com/yz7bfh4p

🤝 Co-funded by the European Commission and implemented together with our partner Factum

ශ්‍රී ලංකාවේ මැතිවරණ කතිකාව හැඩගස්වන සෑම හඬක්ම දේශපාලන පක්ෂයකට හෝ ප්‍රවෘත්ති මැදිරියකට අයත් නොවේ.

අපගේ නවතම අධ්‍යයනය මගින් මෙම හඬවල් සැබවින්ම කාගේද යන්න විමර්ශනය කෙරේ. මෙහිදී 2024 ශ්‍රී ලංකා ජනාධිපතිවරණයට අදාළව Facebook) සහ YouTube ඔස්සේ පළ වූ අන්තර්ගතයන් 4,219 ක් අපගේ සමාලෝචනයට ලක් වූ අතර, එහිදී පළ කිරීම් ප්‍රමාණය, වේදිකා හරහා ඒවායේ ව්‍යාප්තිය සහ අපේක්ෂකයින් පිළිබඳව සාකච්ඡා වී ඇති ආකාරය පරීක්ෂා කෙරිණි.

මෙම සොයාගැනීම් මගින් පැහැදිලි රටා කිහිපයක් පෙන්වා දෙයි.

නිරීක්ෂණය කරන ලද සමස්ත අන්තර්ගතයෙන් 93% කට වඩා වැඩි ප්‍රමාණයක් Facebook හි පළ වී තිබූ අතර, මැතිවරණයට අදාළ සාකච්ඡා සඳහා වන ප්‍රධානතම වේදිකාව ලෙස එහි ස්ථානය මින් තහවුරු වේ. මැතිවරණයට අදාළ සියලුම අන්තර්ගතයන්ගෙන් 81.3% ක් සඳහාම සමාජ මාධ්‍ය බලපෑම්කරුවන් (influencers) දස දෙනෙකු වගකිව යුතු වූ අතර, මින් පෙන්නුම් කරන්නේ සුළු ගිණුම් සංඛ්‍යාවක් අත මෙම අන්තර්ගතයන් ඉහළ මට්ටමකින් සංකේන්ද්‍රණය වී ඇති බවයි. නීතියෙන් නියම කර ඇති මැතිවරණ නිහඬ කාලසීමාව තුළදී පවා මැතිවරණයට අදාළ අන්තර්ගතයන් කැපී පෙනෙන ලෙස පළ වී තිබුණි. එම කාලසීමාව තුළ පළ වූ මුළු මැතිවරණ පෝස්ටු වලින් තුනෙන් එකකට ආසන්න ප්‍රමාණයක් ගිණුම් දෙකකට පමණක් අයත් විය.

එවකට අපේක්ෂකයින් වූ අනුර කුමාර දිසානායක, රනිල් වික්‍රමසිංහ, සජිත් ප්‍රේමදාස සහ නාමල් රාජපක්ෂ යන අයව ප්‍රවර්ධනය කිරීම්, විවේචන සහ අදහස් දැක්වීම් හරහා නිරීක්ෂණ කාලසීමාව පුරාවටම අඛණ්ඩව සාකච්ඡාවට බඳුන් වී තිබුණි.

මෙම සොයාගැනීම් වලින් සමස්තයක් ලෙස පෙනී යන්නේ, අන්තර්ජාලය හරහා ඡන්දදායකයින් වෙත ළඟා වන්නේ කුමන අපේක්ෂකයින්ද සහ කුමන පණිවිඩද යන්න තීරණය කිරීමේදී සාපේක්ෂව කුඩා සමාජ මාධ්‍ය බලපෑම්කරුවන් කණ්ඩායමක් විසින් සැලකිය යුතු බලපෑමක් සිදු කරන බවයි.

සම්පූර්ණ අධ්‍යයන වාර්තාව මෙතැනින් කියවන්න: https://tinyurl.com/yz7bfh4p

🤝 යුරෝපා සංගමයේ සම-මූල්‍ය දායකත්වයෙන් සහ අපගේ හවුල්කාර ආයතනය වන Factum සමඟ එක්ව ක්‍රියාත්මක කරන ලදි.

இலங்கையின் தேர்தல் களம் குறித்த உரையாடல்களை வடிவமைக்கும் ஒவ்வொரு குரலும் ஒரு அரசியல் கட்சிக்கோ அல்லது செய்திப் பிரிவிற்கோ சொந்தமானதல்ல.

இந்தக் குரல்கள் உண்மையில் யாருடையவை என்பதை எமது புதிய ஆய்வு விபரிக்கின்றது. இலங்கையின் 2024 ஜனாதிபதித் தேர்தலின் போது Facebook மற்றும் YouTube தளங்களில் வெளியிடப்பட்ட 4,219 பதிவுகளை நாம் மீளாய்வு செய்தோம். இதில் பதிவுகளின் எண்ணிக்கை, சமூக ஊடக தளங்களில் அவற்றின் பரவல் மற்றும் வேட்பாளர்கள் எவ்வாறு விவாதிக்கப்பட்டார்கள் என்பது ஆராயப்பட்டது.

இந்த ஆய்வின் முடிவுகள் பல தெளிவான போக்குகளைச் சுட்டிக்காட்டுகின்றன.

கண்காணிக்கப்பட்ட அனைத்து உள்ளடக்கங்களிலும் 93% இற்கும் அதிகமானவை Facebook இல் பதிவிடப்பட்டுள்ளன. இது தேர்தல் தொடர்பான விவாதங்களுக்கான முதன்மைத் தளமாக Facebook இருப்பதை உறுதிப்படுத்துகிறது. தேர்தல் தொடர்பான அனைத்து பதிவுகளிலும் 81.3% பதிவுகளுக்கு வெறும் பத்து சமூக வலைத்தள செல்வாக்கு செலுத்துபவர்களே (influencers) பொறுப்பாவார்கள். இது ஒரு சிறிய எண்ணிக்கையிலான கணக்குகளுக்கு மத்தியில் இந்த உள்ளடக்கங்கள் அதிக அளவில் குவிந்துள்ளதைக் காட்டுகிறது. சட்டபூர்வமாக அறிவிக்கப்பட்ட தேர்தல் அமைதிக் காலப்பகுதியிலும் (silence period) தேர்தல் தொடர்பான பதிவுகள் தொடர்ச்சியாகவும் குறிப்பிடத்தக்க வகையிலும் காணப்பட்டன. அக்காலப்பகுதியில் வெளியான அனைத்து தேர்தல் பதிவுகளில் ஏறக்குறைய மூன்றில் ஒரு பங்கு வெறும் இரண்டு கணக்குகளுக்கு மட்டுமே சொந்தமானவையாக இருந்தன.

அப்போதைய வேட்பாளர்களான அனுர குமார திஸாநாயக்க, ரணில் விக்கிரமசிங்க, சஜித் பிரேமதாச மற்றும் நாமல் ராஜபக்ஷ ஆகியோர் கண்காணிப்புக் காலப்பகுதி முழுவதும் விளம்பரப்படுத்தல், விமர்சனங்கள் மற்றும் கருத்துரைகள் ஊடாக தொடர்ந்து விவாதிக்கப்பட்டனர்.

ஒட்டுமொத்தமாக இந்த முடிவுகள், இணையவழி ஊடாக வாக்காளர்களை எந்த வேட்பாளர்களும், செய்திகளும் சென்றடைகின்றன என்பதை தீர்மானிப்பதில், ஒப்பீட்டளவில் ஒரு சிறிய குழுவாக இருக்கும் செல்வாக்கு செலுத்துபவர்கள் (influencers) கணிசமான தாக்கத்தை ஏற்படுத்துகிறார்கள் என்பதைக் காட்டுகின்றன.

முழுமையான ஆய்வறிக்கையை இங்கே வாசிக்கவும்: https://tinyurl.com/yz7bfh4p

🤝 இத்திட்டம் ஐரோப்பிய ஒன்றியத்தின் இணை நிதியுதவியுடன் எமது பங்காளர்களான Factum நிறுவனத்துடன் இணைந்து நடைமுறைப்படுத்தப்பட்டது.

30/07/2026

📢 We're calling for tenders! Looking for an independent evaluator (an individual or a team) to review our project here in Sri Lanka. Could that be you?

Our project, Strengthening Resilience Against Disinformation in Sri Lanka, is co-funded by the European Commission and implemented with Factum. With nearly two years in, we are looking for an external project evaluator (m/f/d) to look at how well the project has delivered on its objectives, working with partners, the Election Commission of Sri Lanka, and communities on the ground.

🗓️ Contract duration: 26 August – 15 November 2026
📍 Work location: Remote, Sri Lanka, with fieldwork in Colombo and beyond

We're looking for someone with:
🎓 Experience evaluating projects on elections, digital democracy, disinformation, hate speech, or capacity building
📚 At least two evaluation or research reports under their belt
🌏 Prior work in Sri Lanka or the region
🗣️ Excellent English and professional knowledge of Tamil and/or Sinhala (required)

✅ Applications close 12 August 2026, 18:00 CET

Full details, required documents, and how to apply: https://tinyurl.com/37zz3bv9

Know someone who fits the brief? Tag them below! 👇

22/07/2026

Your AI Chatbot answer has footnotes. That does not mean it is true.

A citation at the bottom creates an impression of proof. A link, a source, a line such as "according to Wikipedia," and the answer appears verified before anyone reads a word of it.

The real risk with AI chatbots is not fabrication. It is synthesis.

When a chatbot has no clean answer ready, it behaves less like a witness and more like an editor working against a deadline. It draws fragments from several genuine sources and stitches them into a single, confident reply. A government page here, a Wikipedia entry there, a headline from months ago. Each fragment may be accurate on its own. The stitching is where the error takes hold. A detail from last year quietly joins a detail from this year, and the seam vanishes under the final polish.

Nothing about the finished answer looks invented. It carries footnotes. It reads as careful research.

That is precisely what makes it convincing.

A citation confirms that a source exists. It says nothing about whether the synthesis built from it is sound.

We examined this closely in a recent study on chatbots and Sri Lanka's 2025 local elections. Link for readers who wish to go further. https://tinyurl.com/4ujp4hvh

🤝 Co-funded by the European Commission and implemented together with our partner Factum

ඔබේ AI Chatbot පිළිතුරෙහි පාදක සටහන් තිබිය හැකිය. එයින් අදහස් වන්නේ එය සත්‍යයක් බව නොවේ.

පහළින් දක්වා ඇති උපුටා දැක්වීමක් මගින් එය සත්‍ය බව ඔප්පු කර ඇති බවට හැඟීමක් ඇති කරයි. සබැඳියක්, මූලාශ්‍රයක් හෝ "විකිපීඩියාවට අනුව" වැනි වාක්‍ය ඛණ්ඩයක් තිබූ පමණින්, කිසිවෙකු එය කියවීමට පෙරම එම පිළිතුර තහවුරු කර ඇති බවක් පෙනෙන්නට සලස්වයි.

AI Chatbot සම්බන්ධයෙන් ඇති සැබෑ අවදානම ව්‍යාජ තොරතුරු ගෙතීම නොවේ. එය තොරතුරු එකලස් කිරීම හෙවත් සංස්ලේෂණය කිරීමයි.

Chatbot එකක් සතුව ලබා දීමට පැහැදිලි පිළිතුරක් නොමැති විට, එය සාක්ෂිකරුවෙකු මෙන් නිහඬව සිටීමෙන් හෝ පිළිතුරු දීමෙන් වළකින අතර, ඒ වෙනුවට කාල සීමාවකට යටත්ව වැඩ කරන සංස්කාරකවරයෙකු මෙන් ක්‍රියා කරයි. එය සත්‍ය මූලාශ්‍ර කිහිපයකින් කොටස් ලබාගෙන ඒවා එකම විශ්වාසදායක පිළිතුරක් බවට ගලපයි. එක් තැනකින් රජයේ වෙබ් පිටුවක්, තවත් තැනකින් විකිපීඩියා සටහනක්, මාස ගණනකට පෙර පළවූ ප්‍රවෘත්ති සිරස්තලයක් ආදී වශයෙනි. මේ සෑම කොටසක්ම වෙන වෙනම ගත් කල නිවැරදි විය හැකිය. නමුත් වැරැද්ද සිදුවන්නේ ඒවා එකට මූට්ටු කිරීමේදීය. පසුගිය වසරේ තොරතුරක් නිශ්ශබ්දවම මෙම වසරේ තොරතුරක් සමඟ එකතු වන අතර, අවසාන ඔපදැමීමෙන් පසු එම මූට්ටු කළ ස්ථානය නොපෙනී යයි.

මෙම අවසන් පිළිතුරෙහි කිසිවක් අලුතින් ගෙතූ බවක් නොපෙනේ. එහි පාදක සටහන් ඇත. එය ප්‍රවේශමෙන් සිදු කළ පර්යේෂණයක් සේ කියවිය හැකිය.

එය ඒත්තු ගැන්වෙනසුලු වීමට හේතුව හරියටම එයයි.

උපුටා දැක්වීමකින් තහවුරු වන්නේ මූලාශ්‍රයක් පවතින බව පමණි. එයින් ගොඩනගා ඇති සංස්ලේෂණය කෙතරම් දුරට නිවැරදිද යන්න ගැන ඉන් කිසිවක් නොකියවෙයි.

AI Chatbot සහ ශ්‍රී ලංකාවේ 2025 පළාත් පාලන මැතිවරණය පිළිබඳව මෑතකදී සිදු කළ අධ්‍යයනයකදී අපි මේ ගැන සමීපව පරීක්ෂා කළෙමු. තවදුරටත් කියවීමට කැමති පාඨකයන් සඳහා සබැඳිය මෙන්න. https://tinyurl.com/4ujp4hvh

🤝 යුරෝපා සංගමයේ සම-මූල්‍ය දායකත්වයෙන් සහ Factum සමඟ එක්ව ක්‍රියාත්මක කරන ලදි.

உங்களது AI Chatbot பதிலில் அடிக்குறிப்புகள் (Footnotes) இருக்கலாம். அதற்காக அது உண்மையானது என்று அர்த்தமல்ல.

கீழே கொடுக்கப்பட்டுள்ள மேற்கோள் (Citation) அது நிரூபிக்கப்பட்ட ஒன்று போன்ற ஒரு தோற்றத்தை உருவாக்குகிறது. ஒரு இணைப்பு, ஒரு மூலம் அல்லது "விக்கிப்பீடியாவின் படி" போன்ற ஒரு வாசகம் இருந்தால், யாராவது அதைப் படிப்பதற்கு முன்பே அந்தப் பதில் உறுதிப்படுத்தப்பட்டது போன்ற பிம்பத்தை ஏற்படுத்துகிறது.

AI Chatbot தொடர்பான உண்மையான ஆபத்து என்பது தவறான தகவல்களைப் புனைவது (Fabrication) அல்ல. அது தகவல்களைத் தொகுப்பது அல்லது ஒருங்கிணைப்பது ஆகும் (Synthesis).

ஒரு Chatbot-இடம் வழங்குவதற்கு தெளிவான பதில் இல்லாதபோது, அது ஒரு சாட்சியைப் போல மௌனமாக இருப்பதையோ அல்லது பதிலளிப்பதைத் தவிர்ப்பதையோ விடுத்து, அதற்குப் பதிலாக ஒரு குறிப்பிட்ட கால வரையறைக்குள் வேலை செய்யும் ஒரு பதிப்பாசிரியரைப் போலச் செயல்படுகிறது. அது பல உண்மையான தகவல் மூலங்களிலிருந்து பகுதிகளை எடுத்து, அவற்றை ஒரே நம்பகமான பதிலாக இணைக்கிறது. ஒருபுறம் அரசாங்க இணையத்தளப் பக்கம், மறுபுறம் விக்கிப்பீடியா பதிவு, சில மாதங்களுக்கு முன்னரான ஒரு செய்தித் தலைப்பு என பலவாறாக அது அமைகிறது. இந்த ஒவ்வொரு பகுதியும் தனித்தனியாகப் பார்க்கும்போது சரியானதாக இருக்கலாம். ஆனால் தவறுகள் நடப்பது அவற்றை ஒன்றாக இணைக்கும்போதுதான். கடந்த வருடத் தகவல் ஒன்று, இந்த வருடத் தகவலுடன் சத்தமில்லாமல் இணைந்து கொள்கிறது, இறுதி மெருகூட்டலுக்குப் பிறகு அந்த இணைப்புப் பொருத்தம் தெரியாமல் மறைந்துவிடுகிறது.

இந்த இறுதிப் பதிலில் புதிதாகப் புனையப்பட்டதாக எதுவும் தெரிவதில்லை. அதில் அடிக்குறிப்புகள் உள்ளன. அது கவனமாகச் செய்யப்பட்ட ஒரு ஆராய்ச்சி போல் வாசிக்க முடிகிறது.
அது அத்தனை நம்பகத்தன்மையுடன் இருப்பதற்குக் காரணமும் அதுவேதான்.

ஒரு மேற்கோள் மூலமாக ஒரு மூலம் (Source) இருக்கிறது என்பது மட்டுமே உறுதிப்படுத்தப்படுகிறது. அதிலிருந்து உருவாக்கப்பட்ட தொகுப்பு எந்தளவுக்குச் சரியானது என்பது பற்றி அது எதுவும் கூறுவதில்லை.

AI Chatbot மற்றும் இலங்கையின் 2025 உள்ளூராட்சி மன்றத் தேர்தல்கள் குறித்து சமீபத்தில் மேற்கொண்ட ஆய்வில் நாங்கள் இதனை உன்னிப்பாகப் பரிசீலித்தோம். இது தொடர்பில் மேலும் வாசிக்க விரும்பும் வாசகர்களுக்கான இணைப்பு இதோ. https://tinyurl.com/4ujp4hvh

🤝 இத்திட்டம் ஐரோப்பிய ஒன்றியத்தின் இணை நிதியுதவியுடன் எமது பங்காளர்களான Factum நிறுவனத்துடன் இணைந்து நடைமுறைப்படுத்தப்பட்டது.

09/07/2026

In one of our earlier studies on Germany's 2025 federal election, Google's Gemini refused to answer election-related questions altogether. We considered that the safer, more responsible choice. A refusal beats handing voters something wrong.

In our study focusing on Sri Lanka's 2025 local elections, that changed. Gemini answered every single question we asked it, including who the candidates were, and got some of them wrong.

Same company, a different election, a different approach.

It wasn't only Gemini. We tested four chatbots in total: OpenAI's ChatGPT, Google's Gemini, Microsoft's Copilot, and DeepSeek. Only ChatGPT ever declined to answer, and that happened in just 4.2 per cent of cases, a single small exception across the whole study. Every other question we asked, across all four chatbots, got an answer.

Caution shouldn't depend on which election you're asking about. Every voter deserves the same standard, wherever they live.

See the actual screenshots from our research: https://tinyurl.com/2ev8c6c6

🤝 This post is shared under the project Strengthening Resilience Against Disinformation in Sri Lanka. Co-funded by the European Commission and implemented with Factum.

ජර්මනියේ 2025 ෆෙඩරල් මැතිවරණය ගැන අපි කලින් කරපු අධ්‍යයනයකදී, Google හි Gemini මැතිවරණයට අදාළ ප්‍රශ්නවලට පිළිතුරු දීම මුළුමනින්ම ප්‍රතික්ෂේප කළා. අපි ඒක වඩාත් ආරක්ෂිත සහ වගකීම් සහගත තීරණයක් විදිහට සැලකුවා. ඡන්දදායකයින්ට වැරදි තොරතුරක් දෙනවාට වඩා පිළිතුරු දීම ප්‍රතික්ෂේප කරන එක යහපත් වෙනවා.

ශ්‍රී ලංකාවේ 2025 පළාත් පාලන මැතිවරණය ගැන අවධානය යොමු කරපු අපේ අධ්‍යයනයේදී ඒ තත්ත්වය වෙනස් වුණා. අපේක්ෂකයින් කවුද කියන එකත් ඇතුළුව අපි අහපු හැම ප්‍රශ්නයකටම Gemini පිළිතුරු ලබා දුන්නා. ඒ වගේම ඒ සමහරක් තොරතුරු වැරදි වුණා.

සමාගම එකයි, මැතිවරණය වෙනස්, වැඩ කරපු විදිහත් වෙනස්.

මේක Gemini වලට පමණක් සීමා වුණේ නැහැ. අපි chatbots හතරක් පරීක්ෂා කළා: OpenAI හි ChatGPT, Google හි Gemini, Microsoft හි Copilot සහ DeepSeek. පිළිතුරු දීම ප්‍රතික්ෂේප කළේ ChatGPT පමණයි. ඒකත් මුළු අධ්‍යයනය පුරාවටම දකින්න ලැබුණු එකම කුඩා වෙනස විදිහට අවස්ථා 4.2% කදී පමණක් සිදු වුණා. අනිත් චැට්බොට්ස් හතරෙන්ම අපි අහපු හැම ප්‍රශ්නයකටම පිළිතුරක් ලැබුණා.

අපි අහන්නේ මොන මැතිවරණය ගැනද කියන එක මත දක්වන සැලකිල්ල වෙනස් වෙන්න හොඳ නැහැ. හැම ඡන්දදායකයෙකුටම ඔවුන් කොහේ ජීවත් වුණත්, ඒ හැමෝටම එක හා සමාන ප්‍රමිතියකට අයිතියක් තිබෙනවා.

අපේ පර්යේෂණයේ සොයා ගැනීම් මෙතැනින් බලන්න: https://tinyurl.com/2ev8c6c6

🤝 යුරෝපීය සංගමයේ මුල්‍ය දායකත්වයෙන් සහ Factum ආයතනය මගින් ක්‍රියාත්මක කරන "ශ්‍රී ලංකාව තුළ සාවද්‍ය තොරතුරු වලට එරෙහිව ඔරොත්තු දීමේ හැකියාව ශක්තිමත් කිරීමේ" ව්‍යාපෘතිය යටතේ මේ සටහන බෙදාහැර තිබෙනවා.

ஜேர்மனியின் 2025 மத்தியத் தேர்தல் (Federal election) குறித்து நாங்கள் முன்பு மேற்கொண்ட ஒரு ஆய்வில், தேர்தல் தொடர்பான கேள்விகளுக்குப் பதிலளிப்பதை Google இன் Gemini முற்றிலும் நிராகரித்தது. அதனை மிகவும் பாதுகாப்பான மற்றும் பொறுப்பான ஒரு தீர்மானமாக நாங்கள் கருதினோம். வாக்காளர்களுக்குத் தவறான தகவல்களை வழங்குவதை விட, பதிலளிப்பதை மறுப்பது சிறந்ததாகும்.

இலங்கையின் 2025 உள்ளூராட்சி மன்றத் தேர்தல் தொடர்பில் கவனம் செலுத்திய எமது ஆய்வில் அந்த நிலைமை மாறியது. வேட்பாளர்கள் யார் என்பது உட்பட நாங்கள் கேட்ட ஒவ்வொரு கேள்விக்கும் Gemini பதிலளித்தது. அத்துடன் அவற்றில் சில தகவல்கள் தவறானவையாகவும் இருந்தன.

நிறுவனம் ஒன்றுதான், தேர்தல் வேறு, அதற்கு செயற்பட்ட விதமும் வேறு.

இது Gemini இற்கு மாத்திரம் மட்டுப்படுத்தப்பட்டதாக இருக்கவில்லை. நாங்கள் OpenAI இன் ChatGPT, Google இன் Gemini, Microsoft இன் Copilot மற்றும் DeepSeek உட்பட நான்கு chatbots களைப் பரிசோதித்தோம். இதில் ChatGPT மாத்திரமே பதிலளிப்பதை நிராகரித்தது. அதுவும் முழு ஆய்விலும் காணப்பட்ட ஒரு சிறிய வித்தியாசமாக 4.2%மான சந்தர்ப்பங்களில் மட்டுமே நிகழ்ந்தது. நாங்கள் கேட்ட ஏனைய அனைத்து கேள்விகளுக்கும் இந்த நான்கு chatbotsகளிடமிருந்தும் எங்களுக்குப் பதில் கிடைத்தது.

நாம் எந்தத் தேர்தலைப் பற்றிக் கேட்கிறோம் என்பதன் அடிப்படையில் காட்டப்படும் கவனம் மாறுபடக்கூடாது. ஒவ்வொரு வாக்காளரும் எங்கு வாழ்ந்தாலும், அவர்கள் அனைவருக்கும் ஒரே மாதிரியான தரத்தைப் பெறுவதற்கான உரிமை உள்ளது.

எமது பரிசோதனையின் உண்மையான திரைத்துணுக்குகளை (screenshots) இங்கே பார்க்கவும்: https://tinyurl.com/2ev8c6c6

🤝 ஐரோப்பிய ஒன்றியத்தின் நிதியுதவியுடன் Factum நிறுவனத்தால் நடைமுறைப்படுத்தப்படும் "இலங்கையில் தவறான தகவல்களுக்கு எதிரான பின்னடைவை வலுப்படுத்துதல்" எனும் செயற்றிட்டத்தின் கீழ் இந்தப் பதிவு பகிரப்பட்டுள்ளது.

29/06/2026

When an AI flags a misleading post in English, it draws on decades of training data and well-resourced review teams. When the same post appears in Sinhala or Tamil, it travels further, faster, and with less friction.

This is not speculation. It is documented across consecutive electoral cycles in Sri Lanka, and in election after election across the Global South.

This article in Ceylon Today examines why the AI systems governing what stays up and what comes down on platforms like Meta, Google, and X were built, overwhelmingly, for English, and what that structural gap means for Sri Lanka's 22 million people.

https://ceylontoday.lk/2026/06/21/the-languages-ai-ignores-are-putting-democracy-at-risk/

The article draws on a body of research that includes a DRI and Factum study on GenAI risks to Sri Lanka's electoral integrity, which rated platform self-regulation a maximum risk score, pointing specifically to the delayed detection and removal of harmful content in Sinhala- and Tamil-language contexts as one of the country's key electoral vulnerabilities. The 2024 elections showed pockets of improvement. But managing a structural problem more effectively is not the same as resolving it.

Voters in Sri Lanka deserve the same basic protection from AI-driven manipulation as voters anywhere else. Whether platforms, regulators, and the international community choose to treat that with urgency is, at this point, the only question left to answer.

The article was published as part of the Strengthening Resilience Against Disinformation in Sri Lanka project, co-funded by the European Commission and implemented with Factum. The project aims to strengthen Sri Lanka’s democratic and electoral frameworks and protect human rights by promoting inclusivity and responsible political discourse online.

By Chamal Weerakkody For the world’s largest social media platforms, Sri Lanka is not a priority market. That is a business reality, not an accusation. But it carries consequences for democracy that go far beyond advertising revenue.  Those consequences are most visible during elections, and most...

17/06/2026

When you read or share political content in Sinhala or Tamil online, there is a good chance that no AI system is checking it for harm.

The platforms we use every day were built to moderate content in English. For Sinhala and Tamil, the tools are far weaker. During Sri Lanka's 2024 elections, harmful content in local languages passed through platforms largely unchecked, while the same content in English would have had a higher chance of being caught.

Curious about what’s slipping through the cracks? Read our deep-dive into this algorithmic blind spot, https://tinyurl.com/bddcnkwu. UN data recorded a 113 percent rise in harmful speech during the November 2024 Parliamentary Election, much of it targeting ethnic and religious minorities.

This is not unique to Sri Lanka. Countries across Africa and Southeast Asia face the same problem. But unlike the European Union and India, which are now introducing laws that require platforms to moderate content fairly across languages, Sri Lanka has no comparable framework in place.

🤝 This post is shared under the project Strengthening Resilience Against Disinformation in Sri Lanka. Co-funded by the European Commission and implemented with Factum.

අන්තර්ජාලය ඔස්සේ ඔබ සිංහල හෝ දෙමළ භාෂාවෙන් දේශපාලනික තොරතුරු කියවන විට හෝ බෙදාගන්නා විට, ඒවයේ හානිකර බව කිසිදු කෘත්‍රිම බුද්ධි (AI) පද්ධතියකින් පරීක්ෂාවට ලක් නොවීමට බොහෝදුරට ඉඩ තිබෙනවා.

අපි දිනපතා භාවිත කරන සමාජ මාධ්‍ය වේදිකා නිර්මාණය කරලා තියෙන්නේ ඉංග්‍රීසි භාෂාවේ අන්තර්ගතයන් නියාමනය කරන්නයි. සිංහල සහ දෙමළ භාෂාවන් සඳහා මේ මෙවලම් පවතින්නේ ඉතාම දුර්වල මට්ටමක. ශ්‍රී ලංකාවේ 2024 මැතිවරණ සමයේදී දේශීය භාෂාවලින් පළ වූ හානිකර දේවල් බොහෝදුරට කිසිදු පරීක්ෂාවකින් තොරව මේ වේදිකා හරහා පැතිර ගියා. හැබැයි ඒ දේවල්ම ඉංග්‍රීසි භාෂාවෙන් පළ වුණා නම්, ඒවා හඳුනාගෙන ඉවත් කරන්න තිබුණු ඉඩකඩ ගොඩක් වැඩියි.

මේ හැංගිලා තියෙන ඇත්ත තත්ත්වය ගැන අපි කරපු සවිස්තරාත්මක ගවේෂණය මේ දිගුවෙන් (https://tinyurl.com/bddcnkwu) ඔබටත් කියවන්න පුළුවන්. එක්සත් ජාතීන්ගේ දත්ත පෙන්වා දෙන විදිහට, 2024 නොවැම්බර් මහ මැතිවරණ සමයේදී හානිකර ප්‍රකාශවල 113% ක විශාල වර්ධනයක් වාර්තා වෙලා තිබෙනවා. එයින් වැඩි ප්‍රමාණයක් ඉලක්ක කර තිබුණේ වාර්ගික සහ ආගමික සුළුතර කණ්ඩායම්.

මේක ශ්‍රී ලංකාවට විතරක් සීමා වුණු ප්‍රශ්නයක් නෙවෙයි. අප්‍රිකාවේ සහ ගිනිකොනදිග ආසියාවේ රටවල් පවා මුහුණ දෙන්නේ මේ ගැටලුවටමයි. නමුත්, හැම භාෂාවක් පුරාම සාධාරණ විදිහට අන්තර්ගතය නියාමනය කරන්න සමාජ මාධ්‍ය වේදිකාවලට බල කෙරෙන නීති මේ වෙද්දී හඳුන්වා දෙමින් ඉන්න යුරෝපා සංගමය සහ ඉන්දියාව වගේ නෙවෙයි, ශ්‍රී ලංකාව තුළ තවමත් ඒ වගේ ක්‍රමවත් නෛතික රාමුවක් ක්‍රියාත්මක වෙන්නේ නැහැ.

🤝 මෙම සටහන බෙදා හරින්නේ 'ශ්‍රී ලංකාව තුළ සාවද්‍ය තොරතුරු වලට එරෙහිව ඔරොත්තු දීමේ හැකියාව ශක්තිමත් කිරීම' ව්‍යාපෘතිය යටතේයි. මෙය යුරෝපා සංගමයේ සම-මූල්‍ය දායකත්වයෙන් සහ Factum ආයතනයේ මැදිහත්වීමෙන් ක්‍රියාත්මක කරනවා.

இணையத்தில் நீங்கள் சிங்களம் அல்லது தமிழ் மொழியில் அரசியல் தொடர்பான தகவல்களை வாசிக்கும்போதோ அல்லது பகிரும்போதோ, அவற்றின் தீங்கு விளைவிக்கும் தன்மையை எந்தவொரு செயற்கை நுண்ணறிவு (AI) முறைமையும் சரிபார்க்காமல் இருக்க அதிக வாய்ப்புள்ளது.

நாம் தினமும் பயன்படுத்தும் சமூக ஊடகத் தளங்கள் பெரும்பாலும் ஆங்கில மொழியிலான உள்ளடக்கங்களை ஒழுங்குபடுத்துவதற்காகவே வடிவமைக்கப்பட்டுள்ளன. சிங்களம் மற்றும் தமிழ் மொழிகளைப் பொறுத்தவரை, இந்த தொழில்நுட்பக் கருவிகள் மிகவும் பலவீனமான நிலையிலேயே காணப்படுகின்றன. இலங்கையின் 2024 தேர்தல் காலத்தில், உள்ளூர் மொழிகளில் வெளியான தீங்கு விளைவிக்கும் பதிவுகள் பெரும்பாலும் எந்தவித தணிக்கையுமின்றி இந்த தளங்கள் ஊடாக பரவின. ஆனால், அதே பதிவுகள் ஆங்கிலத்தில் வெளியாகியிருந்தால், அவை கண்டறியப்பட்டு நீக்கப்படுவதற்கான வாய்ப்புகள் மிக அதிகமாக இருந்திருக்கும்.

மறைந்திருக்கும் இந்த உண்மை நிலை குறித்து நாங்கள் மேற்கொண்ட விரிவான ஆய்வை இந்த இணைப்பின் மூலம் (https://tinyurl.com/bddcnkwu) நீங்களும் வாசிக்கலாம். ஐக்கிய நாடுகள் சபையின் தரவுகளின்படி, 2024 நவம்பர் நாடாளுமன்றத் தேர்தல் காலத்தில் தீங்கு விளைவிக்கும் கருத்து வெளியீட்டில் 113% பாரிய அதிகரிப்பு பதிவாகியுள்ளது. அவற்றில் பெரும்பாலானவை இன மற்றும் மத ரீதியான சிறுபான்மையினரை இலக்காகக் கொண்டே அமைந்திருந்தன.

இது இலங்கைக்கு மாத்திரம் மட்டுப்படுத்தப்பட்ட ஒரு பிரச்சினையல்ல. ஆப்பிரிக்கா மற்றும் தென்கிழக்காசிய நாடுகளும் இதே சவாலை எதிர்கொள்கின்றன. ஆனால், அனைத்து மொழிகளிலும் உள்ளடக்கங்களை நியாயமான முறையில் ஒழுங்குபடுத்த சமூக ஊடகத் தளங்களைக் கட்டாயப்படுத்தும் சட்டங்களை தற்போது அறிமுகப்படுத்தி வரும் ஐரோப்பிய ஒன்றியம் மற்றும் இந்தியாவைப் போலல்லாமல், இலங்கையில் இன்னமும் முறையான சட்டக் கட்டமைப்பு எதுவும் நடைமுறையில் இல்லை.

🤝 இந்தப் பதிவு 'இலங்கையில் தவறான தகவல்களுக்கு எதிரான மீள்திறனை வலுப்படுத்துதல்' (Strengthening Resilience Against Disinformation in Sri Lanka) எனும் செயற்றிட்டத்தின் கீழ் பகிரப்படுகிறது. இது ஐரோப்பிய ஒன்றியத்தின் இணை நிதியுதவியுடனும் Factum நிறுவனத்தின் பங்களிப்புடனும் நடைமுறைப்படுத்தப்படுகிறது.

11/06/2026

What happens when voters trust a fake account more than an official source? Or when an AI chatbot confidently provides the wrong answer about an election?

This article explores how fake political accounts and inaccurate AI-generated information shaped the information environment during Sri Lanka's 2025 Local Government Elections.

Drawing on findings from DRI's research, the article examines,
🔹 Unofficial political accounts posing as authentic voices online
🔹 How major AI chatbots responded to election-related questions
🔹 The risks these trends pose to voters and democratic processes
🔹 Why stronger accountability from digital platforms matters

Read the full article: https://ceylontoday.lk/2026/05/24/how-fake-accounts-and-confused-chatbots-are-quietly-rewriting-elections/

The analysis draws on insights from DRI's studies on political impersonation and AI-generated election misinformation in Sri Lanka.

The article was published as part of the Strengthening Resilience Against Disinformation in Sri Lanka project, co-funded by the European Commission and implemented with Factum. The project aims to reinforce Sri Lanka’s democratic and electoral frameworks and protect human rights by promoting inclusivity and responsible political discourse online.

By Chamal Weerakkody Imagine you are a first-time voter. Sri Lanka’s Local Government Elections, delayed for nearly seven years, have finally arrived. You pick up your phone a few days before polling day. You scroll TikTok, watching video after video from what look like official party pages: the l...