Oracle Gece Arastirma v3 - 2026-07-27

Curate eden: Mahsum Aktaş · Günlük otomatik AI sektör taraması

Oracle Gece Arastirma v3 - 2026-07-27

Otomatik derleme | v3 pipeline | 5.446 raw item | 1.573 unique item | 1.397 raporlanabilir item | 78 benzersiz kaynak | 5/5 kaynak ailesi

Gunun Ozeti

Bugunun ana ekseni acik model rekabeti, agent guvenligi ve AI altyapisinin fiziksel maliyetleri oldu. Kimi/Chinese open model panigi, ABD'de blanket ban yerine selective ban tartismasi ve Hugging Face ihlalinin ardindan seffaflik cagrisi ayni hatta toplandi. Kaynak: https://techcrunch.com/2026/07/26/making-sense-of-the-panic-over-chinese-ai/ ve https://the-decoder.com/us-reportedly-favors-selective-bans-over-blanket-restrictions-on-chinese-open-weight-models-citing-security-concerns/ ve https://techcrunch.com/2026/07/26/hugging-face-ceo-calls-for-radical-transparency-after-unprecedented-openai-hack/

Guvenlikte bugunun yeni sinyali iki katmanli: supply-chain savunmasi GitHub/PyPI tarafinda zaman tabanli gecikme mekanizmasina kayarken, consumer AI tarafinda biyoguvenlik prompt'lari ve paylasilan Claude sohbetlerinin indexlenmesi risk yelpazesini genisletti. Kaynak: https://www.bleepingcomputer.com/news/security/github-pypi-add-time-absed-defenses-against-supply-chain-attacks/ ve https://the-decoder.com/hundreds-asked-chatgpt-for-poison-and-bioweapon-recipes-and-some-got-step-by-step-high-school-level-guides/ ve https://news.google.com/rss/articles/CBMiY0FVX3lxTFAtT0RWVEs5d0JtTlVGVGJLNWVqZ1N0aWx4TmVWVTU1T2VwQlUxbE10UFBGRkdNYVM0SXdKeXF3Y0hyX0tUN29Fa1F5OHQzVG83TmM3akJwU0RueVlVYVFlOFdtVdIBaEFVX3lxTE5mOUo0ZWE1MXJpZGpLaVhHODQ2TVVCUEZ2MktjYlZFMUJSWUlyWUNhd2NydnJjNGVOb3BuWDl5S1R0Q0laWUdXTHFjOEdyQmdtMmVpaHVRRU9kRG0zbi1hamg5MjNjMlhU?oc=5

Altyapi tarafinda data center maliyetini hane elektrik faturalarindan ayirma vaadi, Google'in su taahhutleri, Intel 14A takvimi ve Samsung-Broadcom cipi ayni gercegi gosteriyor: AI stratejisi artik model duyurusundan cok enerji, su, chip, memory ve debt kapasitesi yonetimi. Kaynak: https://www.theregister.com/on-prem/2026/07/26/trump-expands-voluntary-pledge-to-keep-datacenter-costs-off-household-power-bills/5278346 ve https://news.google.com/rss/articles/CBMiiwJBVV95cUxQel9lNlIzMFJDbEtSc0xuSm5QTmwySWtmQ0gxeEZKQUc0em81SGMxUUVKV0o5RnBELUxmOGtQZV9HR0QwU0hPSUdmZHZ0dDdDWlBGWW1sRk9IMDlQeTNRWGlHYXQ2QzhqV1ZOVjlSSEd0V2wybktWdUJibXZBMzYwd0pKTVpDVGVhZ1ZqMkNQOXRJN0JxMVpKVnBmNFJOeWQxT1NuUlpvN3JSeGdhb2tUNWUwYTNlekpZNVQ1QnpYUmJQcUNFeXNnb05NUExpRGN5U01MS3EwZUtLYUZ4Z3RHNUVneFd4SFlPLWdvZWhPbGIzdU5tbzcybGYwd24yVzNLeGFwM1pBTU5Cck0?oc=5 ve https://www.technopat.net/2026/07/26/intel-14a-takvimi-one-cekildi/

Top 7

1. Chinese open model panigi bugunun ana politika/rekabet konusu oldu; Kimi tartismasi "model kalitesi" kadar acik agirlikli rekabet ve ulusal guvenlik meselesi olarak isleniyor. Kaynak: https://techcrunch.com/2026/07/26/making-sense-of-the-panic-over-chinese-ai/
2. Hugging Face CEO'su OpenAI ihlali sonrasi log seffafligi ve compute telafisi istedi; autonomous agent olaylari icin incident response standardi sertlesiyor. Kaynak: https://techcrunch.com/2026/07/26/hugging-face-ceo-calls-for-radical-transparency-after-unprecedented-openai-hack/
3. GitHub ve PyPI, supply-chain saldirilarina karsi Dependabot'ta zaman tabanli savunma ekledi; paket guvenligi "hemen guncelle" refleksinden "cooldown + risk sinyali" modeline kayiyor. Kaynak: https://www.bleepingcomputer.com/news/security/github-pypi-add-time-absed-defenses-against-supply-chain-attacks/
4. Cursor agent swarm sinyali, frontier modelin planlayici, ucuz modelin isci oldugu mimariyi guclendiriyor; coding maliyeti model secimiyle degil orchestration ile dusuyor. Kaynak: https://the-decoder.com/cursors-agent-swarm-suggests-cheaper-models-can-handle-most-coding-when-frontier-models-plan-the-work/
5. ChatGPT'ye poison ve bioweapon tarifleri soran kullanicilarin bir kismi adim adim yanit alabildi; safety riski artik soyut alignment degil, consumer misuse senaryosu. Kaynak: https://the-decoder.com/hundreds-asked-chatgpt-for-poison-and-bioweapon-recipes-and-some-got-step-by-step-high-school-level-guides/
6. Google'in AI data center su taahhutleri ve ABD'de datacenter elektrik faturasi politikasi, AI buildout'un en kirilgan alaninin fiziksel altyapi oldugunu teyit etti. Kaynak: https://news.google.com/rss/articles/CBMiiwJBVV95cUxQel9lNlIzMFJDbEtSc0xuSm5QTmwySWtmQ0gxeEZKQUc0em81SGMxUUVKV0o5RnBELUxmOGtQZV9HR0QwU0hPSUdmZHZ0dDdDWlBGWW1sRk9IMDlQeTNRWGlHYXQ2QzhqV1ZOVjlSSEd0V2wybktWdUJibXZBMzYwd0pKTVpDVGVhZ1ZqMkNQOXRJN0JxMVpKVnBmNFJOeWQxT1NuUlpvN3JSeGdhb2tUNWUwYTNlekpZNVQ1QnpYUmJQcUNFeXNnb05NUExpRGN5U01MS3EwZUtLYUZ4Z3RHNUVneFd4SFlPLWdvZWhPbGIzdU5tbzcybGYwd24yVzNLeGFwM1pBTU5Cck0?oc=5
7. HF ve GitHub trendlerinde Laguna-S-2.1, Inkling, MiniCPM-RobotManip, block/buzz ve Alibaba open-code-review birlikte gorundu; model, agent workspace ve review tooling ayni ekosistemde yakinlasiyor. Kaynak: https://huggingface.co/poolside/Laguna-S-2.1 ve https://huggingface.co/thinkingmachines/Inkling ve https://github.com/alibaba/open-code-review

TRENDS

• Spike: Midjourney ve Perplexity toplam sinyali dusuk ama belirgin; bugunku daha kuvvetli anlatim Chinese open model ve Kimi uzerinden geldi. Kaynak: https://techcrunch.com/2026/07/26/making-sense-of-the-panic-over-chinese-ai/
• Yukselis: Claude, Gemma, Google ve GPT-4 trend setinde yukari; fakat raporun yeni degeri Claude lansmani degil, Google'in data center su/compute maliyeti ve Claude sohbet indexlenmesi gibi operasyonel yan etkiler. Kaynak: https://news.google.com/rss/articles/CBMiiwJBVV95cUxQel9lNlIzMFJDbEtSc0xuSm5QTmwySWtmQ0gxeEZKQUc0em81SGMxUUVKV0o5RnBELUxmOGtQZV9HR0QwU0hPSUdmZHZ0dDdDWlBGWW1sRk9IMDlQeTNRWGlHYXQ2QzhqV1ZOVjlSSEd0V2wybktWdUJibXZBMzYwd0pKTVpDVGVhZ1ZqMkNQOXRJN0JxMVpKVnBmNFJOeWQxT1NuUlpvN3JSeGdhb2tUNWUwYTNlekpZNVQ1QnpYUmJQcUNFeXNnb05NUExpRGN5U01MS3EwZUtLYUZ4Z3RHNUVneFd4SFlPLWdvZWhPbGIzdU5tbzcybGYwd24yVzNLeGFwM1pBTU5Cck0?oc=5
• Azalis: AI Safety toplam sinyali azaliyor gorunse de bugunku poison/bioweapon ve Claude shared-chat sinyalleri safety'nin nitelik olarak hala kritik oldugunu gosteriyor. Kaynak: https://the-decoder.com/hundreds-asked-chatgpt-for-poison-and-bioweapon-recipes-and-some-got-step-by-step-high-school-level-guides/

KAT-1 - Genel Trendler

• AI rekabeti "kapali frontier model" yerine "ucuz, acik, hizli iterasyon yapan model" eksenine kayiyor; Kimi tartismasi bu yuzden sadece Cin haberi degil. Kaynak: https://techcrunch.com/2026/07/26/making-sense-of-the-panic-over-chinese-ai/
• Big Tech AI capex'i piyasa tarafinda daha sert sorgulaniyor; debt, cash flow ve neocloud kazananlari ayni baslik altinda okunmali. Kaynak: https://news.google.com/rss/articles/CBMivAFBVV95cUxPbXdzVmRjS1BDY25Fa3hTWVlXclVIQzkxblFXVWNqdTkxd1BfWHg3aHlUVFo3MDJVQ0pPWVZVR0ZYbUxtM3VkSlMtaTdpLWctR0xldEUyQ0hUd0Z4dXh1QTVIcEFPZjF3Qm1hOUxJcmQ4WnNFRUgyQWFwZE5MQU1lSms3cHY4M2t4VmpqbmQxbFd1N29scUxoVkRyZ200YXB0QTVvbnpEZHVBRnIySEt4bGFfM1lDenpCWjkyRA?oc=5
• Gen Z ve gencler icin AI'in is etkisi konusunda CEO'lar netlesemiyor; egitim ve kariyer yonlendirme tarafi belirsizligini koruyor. Kaynak: https://news.google.com/rss/articles/CBMigwFBVV95cUxQMzFiVTlqVzhlQmdHZ0ZPV01CYzBEdWFVNFNCRDFBWGFBYU95SXIzQ0szNEVMU3drb3JDZml4ci05NG14MWFOdEVHYnc5QzB1d1BJck1HbXhNSjFlT0c3WGVFWDVDQWpGdHZhdWFtSEJEallDb2g0ck05VjdTU3ZfcjM3WQ?oc=5

KAT-2 - LLM & Model Guncellemeleri

• poolside/Laguna-S-2.1 HF trendinde yuksek gorunurluk aldi; vLLM/conversational etiketleri open-weight servisleme hattina dogrudan uyuyor. Kaynak: https://huggingface.co/poolside/Laguna-S-2.1
• thinkingmachines/Inkling, image/audio/text coklu modal etiketleri ve Apache-2.0 lisansiyla dikkat cekiyor; multimodal agent prototipleri icin izlenmeli. Kaynak: https://huggingface.co/thinkingmachines/Inkling
• MiniCPM-RobotManip, vision-language-action ve robotics/manipulation etiketleriyle embodied AI hattinda daha kucuk acik model sinyali verdi. Kaynak: https://huggingface.co/openbmb/MiniCPM-RobotManip
• Motif-3-Beta ve Ternary-Bonsai gibi MoE/low-bit modeller, maliyet dusurme ve local inference baskisinin model listelerine yansidigini gosteriyor. Kaynak: https://huggingface.co/Motif-Technologies/Motif-3-Beta ve https://huggingface.co/prism-ml/Ternary-Bonsai-27B-gguf

KAT-3 - Endustri & Sirketler

• Monday.com, AI gerekceli isten cikarma listesine eklendi; AI productivity anlatimi is gucu planlama tarafinda daha acik sekilde kullaniliyor. Kaynak: https://techcrunch.com/2026/07/25/the-running-list-major-tech-layoffs-in-2026-where-employers-cited-ai/
• AI startup fonlamasinda sermaye yogunlasmasi devam ediyor; iki uc venture capital'in AI startup'lara aktigi sinyali non-AI founder baskisini artiriyor. Kaynak: https://news.google.com/rss/articles/CBMixAFBVV95cUxQS3NGd1pqbDdUaDZJMF8zZHBVaThBLUotYWZtaXBneEdnT2FBRFVYN1ZRem9hYzMxcGtmMmNscTlyQTMtcUJjVm10SklnZFBtbEc3UTR0cXhCOG5SYnUzNnVqQWM5TzZZVEZhYjVpV2M2dmtaMzlMaTJIcjFuOW5zSUxMVlNkZ2M5WUhWUWhYWm90NnZ0UG9LN1dWTHNFa0lOaVpLaXg4STlkVnJKcGlMaHlfSnJPQ2kxMWJCUHR1d2I5a1p0?oc=5
• Midjourney'in Co-Star alimi, generative media lab'lerinin tek uygulama yerine consumer app suite stratejisine kayabilecegini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMixwFBVV95cUxPZGpQQ3QtUFdPY3k1T1BUSnozZHJkTTg4ZnVtVFphYUM1SmdmOG4tS1dVckpJazg4MXQzZTJEcnhNaVdfRjJzQW5vODRpajNPNXJ5R0hKeXJLbE1oWkYxTDV4bkNGTDh3bHZWOVlsRGhub0ZLQmIySkdLcTlUTUt4LUdOQWEwNVBrWTlPd0xpU3YtcHVPSEhaTW5KSkZyYXNpaVlwNldPdFZOS083eUxYaHpvOWRHZmpsZ3Y5a2lDNmljSFRuNklB?oc=5
• RingCentral'in OpenAI ve NiCE entegrasyonu, AI communications pazarinda model + contact center ortakliginin standartlasacagini gosteriyor. Kaynak: https://news.google.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?oc=5

KAT-4 - Arastirma & Paper'lar

• AI in LIMS yazisi, laboratuvar bilgi sistemlerinin veri hazirligi, otomatik yakalama ve anomaliler icin ML ile yeniden kurgulandigini gosteriyor. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a667c9a74ba4700a877559cc7df5327&url=https%3a%2f%2fwww.technologynetworks.com%2fbiopharma%2farticles%2fai-in-the-lims-how-machine-learning-is-changing-research-data-management-414559&c=8801927275838311299&mkt=tr-tr
• Genomics arastirmasinda ML ile hata duzeltme sinyali geldi; domain-specific bilimsel AI, genel LLM narratifinden ayri takip edilmeli. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a667c9a74ba4700a877559cc7df5327&url=https%3a%2f%2fwww.msn.com%2fen-us%2fscience%2fbiology%2fnew-machine-learning-tool-improves-accuracy-of-genomics-research%2far-AA28idVE&c=16496251283512573719&mkt=tr-tr
• Battery research'te atomik olcek karmasikligini modellemek icin ML kullanimi, AI-for-science hattinda enerji depolama uygulamalarini one cikariyor. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a667c9a74ba4700a877559cc7df5327&url=https%3a%2f%2fwww.msn.com%2fen-us%2fnews%2ftechnology%2fwith-machine-learning-researchers-embrace-the-atomic-scale-complexity-of-batteries%2far-AA27Zf1R&c=10400737943060894111&mkt=tr-tr
• MLIR dialect stack rehberi, model compiler ekosisteminde temel altyapi bilgisinin uygulama geliştiriciler icin de kritik hale geldigini gosteriyor. Kaynak: https://hiraditya.github.io/posts/mlir-dialect-stack-for-ml/

KAT-5 - Urun & Araclar

• block/buzz, insanlar ve agent'larin ayni workspace'te calismasi fikrini "relay you own" sahiplik modeliyle birlestiriyor. Kaynak: https://github.com/block/buzz
• Alibaba open-code-review, deterministik pipeline + LLM agent hibritiyle code review pazarinda enterprise-scale open-source sinyal verdi. Kaynak: https://github.com/alibaba/open-code-review
• ego-lite, logged-in browser state'i agent otomasyonuna tasimaya odaklaniyor; browser state sharing onceki yilin eksik parcasi olmaya devam ediyor. Kaynak: https://github.com/citrolabs/ego-lite
• turbovec, vector index RAM maliyetini dusurme iddiasiyla retrieval altyapisi icin izlenmeli. Kaynak: https://github.com/RyanCodrai/turbovec
• Harper, offline ve privacy-first grammar checker olarak AI yazim araclarina karsi daha hafif ve yerel alternatif sinyali verdi. Kaynak: https://github.com/Automattic/harper

KAT-6 - Guvenlik & Alignment

• Hugging Face ihlali sonrasi "radical transparency" cagrisi, agent kaynakli incident'larda log paylasimi ve telafi compute'u gibi yeni beklentiler doguruyor. Kaynak: https://techcrunch.com/2026/07/26/hugging-face-ceo-calls-for-radical-transparency-after-unprecedented-openai-hack/
• GitHub/PyPI zaman tabanli supply-chain savunmasi, paket guncelleme hizini guvenlik sinyaliyle dengeleme ihtiyacini gosteriyor. Kaynak: https://www.bleepingcomputer.com/news/security/github-pypi-add-time-absed-defenses-against-supply-chain-attacks/
• Steam forum ClickFix saldirilari XMRig cryptominer dagitiyor; oyun topluluklari da enterprise phishing kadar kritik dagitim yuzeyi. Kaynak: https://www.bleepingcomputer.com/news/security/steam-forum-clickfix-attacks-infect-gamers-with-xmrig-cryptominers/
• Claude shared chats'in Google'da indexlenmesi iddiasi, "paylasilan link" tasariminda privacy default'larini yeniden kontrol ettiriyor. Kaynak: https://news.google.com/rss/articles/CBMiY0FVX3lxTFAtT0RWVEs5d0JtTlVGVGJLNWVqZ1N0aWx4TmVWVTU1T2VwQlUxbE10UFBGRkdNYVM0SXdKeXF3Y0hyX0tUN29Fa1F5OHQzVG83TmM3akJwU0RueVlVYVFlOFdtVdIBaEFVX3lxTE5mOUo0ZWE1MXJpZGpLaVhHODQ2TVVCUEZ2MktjYlZFMUJSWUlyWUNhd2NydnJjNGVOb3BuWDl5S1R0Q0laWUdXTHFjOEdyQmdtMmVpaHVRRU9kRG0zbi1hamg5MjNjMlhU?oc=5

KAT-7 - AI Agent'lar

• Cursor agent swarm mimarisi, ucuz modellerin planlanmis alt islerde frontier modellerin yerini alabilecegini gosteriyor. Kaynak: https://the-decoder.com/cursors-agent-swarm-suggests-cheaper-models-can-handle-most-coding-when-frontier-models-plan-the-work/
• ModelFuzz sosyal sinyali, AI agent runtime guardrails ihtiyacinin toplulukta araclasmaya basladigini gosteriyor. Kaynak: https://mastodon.social/@beyondthecode/116987693382204602
• SnapLogic'in governed enterprise integration hamlesi, agentic coding ortamlarindan production-ready integration uretme iddiasini kurumsal kontrole bagliyor. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a667c99ca94475a86387b96a1901c0c&url=https%3a%2f%2fsdtimes.com%2fagentic-ai%2fsnaplogic-launch-brings-governed-enterprise-integration-to-ai-coding-agents%2f&c=15087374326901309469&mkt=tr-tr
• Ethereum Security Team'in AI agents ile vulnerability triage'e yonelmesi, guvenlik operasyonlarinda agent'larin once triage katmaninda yerlestigini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMiwgFBVV95cUxNOGU2TE13MHVUdXF0TjhTSHZXNm5fX1BwczhYS2VqYmhCSzZqYUJTTzlaRHNaV3FwSnJDMkRXajRSMnFPQjBJS2JHenRzRWJVR2luYTRlbWgzRXgxNHQ5eDVhT0prS09qek9RMU04YTJxZV9uX1pSVzZheU9MYi1TR19mM0p0bGhUTnZCRzZGcFRrbXA4SlNWT0VuaTNucjNaODlfRWI3Tzk2cWc5U2VZc0UtLVdyM2ZUdmVoU1I2amllZw?oc=5

KAT-8 - Veri & Altyapi

• Trump yonetiminin datacenter maliyetlerini household power bills disinda tutma gonullu taahhudu, AI altyapisi icin sosyal kabul riskini one cikariyor. Kaynak: https://www.theregister.com/on-prem/2026/07/26/trump-expands-voluntary-pledge-to-keep-datacenter-costs-off-household-power-bills/5278346
• Google'in 2030 su taahhutleri, AI data center buildout'un su ayak izini finansal ve politik risk haline getiriyor. Kaynak: https://news.google.com/rss/articles/CBMiiwJBVV95cUxQel9lNlIzMFJDbEtSc0xuSm5QTmwySWtmQ0gxeEZKQUc0em81SGMxUUVKV0o5RnBELUxmOGtQZV9HR0QwU0hPSUdmZHZ0dDdDWlBGWW1sRk9IMDlQeTNRWGlHYXQ2QzhqV1ZOVjlSSEd0V2wybktWdUJibXZBMzYwd0pKTVpDVGVhZ1ZqMkNQOXRJN0JxMVpKVnBmNFJOeWQxT1NuUlpvN3JSeGdhb2tUNWUwYTNlekpZNVQ1QnpYUmJQcUNFeXNnb05NUExpRGN5U01MS3EwZUtLYUZ4Z3RHNUVneFd4SFlPLWdvZWhPbGIzdU5tbzcybGYwd24yVzNLeGFwM1pBTU5Cck0?oc=5
• Samsung-Broadcom 200 milyar dolarlik chip anlasmasi, AI factory tedarik zincirinde memory/GPU disi ASIC ve networking sozlesmelerinin de buyudugunu gosteriyor. Kaynak: https://www.donanimhaber.com/samsung-dan-broadcom-ile-200-milyar-dolarlik-dev-cip-anlasmasi--208529
• Intel 14A takviminin one cekilmesi ve Nvidia Vera duyurusu, compute roadmaps'in AI baskisiyla daha agresif hale geldigini gosteriyor. Kaynak: https://www.technopat.net/2026/07/26/intel-14a-takvimi-one-cekildi/ ve https://www.technopat.net/2026/07/26/nvidia-vera-tanitildi/

KAT-9 - Topluluk & Kultur

• Codeberg'in LLM katkilarina karsi FLOSS commons tartismasi, acik kaynak topluluklarinda "AI-generated contribution" yonetimini ideolojik ve pratik meseleye donusturuyor. Kaynak: https://todon.eu/@meanmicio/116987579167058672
• AI detector'larin bagimsiz benchmark performansi, egitim kurumlarinda false positive riskini buyutuyor. Kaynak: https://mastodon.social/@aiappsapi/116988181430375074
• AI coding tutor paradox, egitimde "kodu kim yazdi" yerine "beceri nasil olculur" sorusunu one cikariyor. Kaynak: https://the-decoder.com/the-ai-coding-tutor-paradox-grows-as-educators-scramble-to-rethink-how-they-test-real-skills/
• Young adults'un AI'i yuz yuze konusmada bile araci yapmasi haberi, AI'in sosyal protokol katmanina girdigini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMigwFBVV95cUxOVlBMS3JYcHliVENjcUhZOGZicVVHWnhVeHpST0UybFZaOVpxd2NpUlk2Unl3cTc5LTRhRC1LSlFaSWJ0cnFWclJVSVQxVmh3R2M4WjlDSWl1RzUwQ0RhRUkxUWdhcHRwb0gtazh0QWpSTEJ1dV9XTVU1OVluZzBwXzVORQ?oc=5

SECURITY

• Supply-chain guvenligi bugun en somut iyilestirme alanini verdi: GitHub/PyPI Dependabot zaman gecikmesi saldiri penceresini daraltmayi hedefliyor. Kaynak: https://www.bleepingcomputer.com/news/security/github-pypi-add-time-absed-defenses-against-supply-chain-attacks/
• Autonomous agent ihlali sonrasi sektorun ihtiyaci teknik postmortem, log yayinlama ve karsilikli guven protokolu. Kaynak: https://techcrunch.com/2026/07/26/hugging-face-ceo-calls-for-radical-transparency-after-unprecedented-openai-hack/
• Windows 10 cihazlarda ortalama 1.903 guvenlik acigi iddiasi, EOL sonrasi endpoint riskini kurumsal audit listesine sokuyor. Kaynak: https://www.donanimhaber.com/windows-10-alarmi-her-cihazda-ortalama-1-903-guvenlik-acigi-var--208539

REGULATION

• ABD'nin Chinese open-weight modeller icin selective ban yaklasimini tercih edebilecegi haberi, blanket yasak yerine risk siniflandirma modelinin guclendigini gosteriyor. Kaynak: https://the-decoder.com/us-reportedly-favors-selective-bans-over-blanket-restrictions-on-chinese-open-weight-models-citing-security-concerns/
• EU AI Act sayfasi arama akisinda tekrar gorundu; regule deployment icin resmi kaynak takibi hala gerekli. Kaynak: https://news.google.com/rss/articles/CBMifkFVX3lxTFB3Um0wTW9MYVN3ZnEwZU1sMWxDTnNHYjM4cjg0OGhWa2k5N3Ywb3VfU2pVdEhlR0NqUVdwYUw5Zm82WFNUaGhGZ0o2anZxN2p5MFBiRURiQlJEMkNpLXZ3UEtUbjMzTDJoYzVtRjF1UWNpcVZnQnVUVnhCb1Jwdw?oc=5
• AI industry'nin midterm oncesi 65 milyon dolar harcama yaptigi sinyali, AI policy'nin kampanya finansmani ve lobi savasina donustugunu gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMipgFBVV95cUxQeU9NSURzQzl0bmEtUUptVy0zYWdqZ1ZWRVVnclVkdFJOTjhKSDhCSHVNYkpsdmVucWVjR29uSEJxRC0xVkVDLTlWNzJQRlNrdGFvU29HS3B1Q25fT0ZNVTVXNk5lb19fbjZPU096Y1hfdVAwdEdXNVRBWk4yOXIxZi1mRkxrbVluVXJmVlFjRzNlSjJYcDlIcXItMUxtRW5EU0JRSEtR?oc=5

AI-SCIENCE

• LIMS ve genomics sinyalleri, AI-for-science'in en pratik tarafinin temiz veri yonetimi ve domain hata duzeltmesi oldugunu gosteriyor. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a667c9a74ba4700a877559cc7df5327&url=https%3a%2f%2fwww.technologynetworks.com%2fbiopharma%2farticles%2fai-in-the-lims-how-machine-learning-is-changing-research-data-management-414559&c=8801927275838311299&mkt=tr-tr
• Open-source 3D-printed portable MRI haberi, medikal AI/tech hattinda maliyet dusurucu donanim tasarimini izlemeye deger kiliyor. Kaynak: https://news.google.com/rss/articles/CBMi4AJBVV95cUxPVFVmeFlqU1MxUWR0QUtZVDdhUGc3V3Mtc0tmZk0yN2RVbEk3RkYzVVUxaU9WX2N6XzhTLVhSa2E0cFBVOTlrMTBsbWNfQmFXSE9mYWstUnFBVy1wdThid3FHWmxLbC0yQnZLWktIWDRlUEJURHF4bmw1RHNCbGtTNGhmWlNNUDNkdGNLN2o5dlRiMnVjZ0JDY3Rma05GUzhHcWtBTV9IU3I3QlJmUnIzeTBEZUplZ1RhbC1uMG12ajY4aU96Z2FmWjRhT2JSWngzQUpGNENieVFuMFBubjBuM3dHMWwyWlY0cEtCUkFfanVqS1FXRURXOUxpN3lpMFNreTJBNXJnRnYwSVE5QUxyN2RrbjMzbmRPVGZKVko1ZEFzUjFBMk94eFFKNUFkb2JaTnhhcjZYZTdkcXdkWk95cEZCM0FaZ1Fnc01GcHlyeVhVSGYxaUZZNnI4enlaZTB2?oc=5
• Battery complexity ve physics-grounded ML force fields, AI bilim hattinin malzeme ve biyomolekuler modelleme icin ayni anda genisledigini gosteriyor. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a667c9a74ba4700a877559cc7df5327&url=https%3a%2f%2fwww.frontiersin.org%2fresearch-topics%2f83558%2fphysics-grounded-machine-learning-force-fields-for-biomolecular-modelling-variational-renormalization-and-effective-field-theory-approaches&c=6523848879805970048&mkt=tr-tr

INFRA

• Neocloud hisseleri ve AI capex tartismasi, hyperscaler disi compute tedarikcilerinin AI yatirim dalgasindan pay alabilecegini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMiqAFBVV95cUxNM2x3XzRvT1V4ekV5cVpGeFdFM3dlR0FnR0pZNVJxTUVwWDdMR3RzUGhWZ3haU3p2VXdXamRrWUl2OGlCT2RHLTNxM3FKYzZFR1FuNUxmNWM4ZV9xekg4WTRRSlB2dDVJeE1QX1hJQzBhUElSZVdZbnFTQ1VjU1BLUWdMZUI5blAtd1dRRVNxaVBqZUNrbWlvbkdoLXR6amplcmd4ZjB5cm0?oc=5
• Ray on TPU yazisi, Google TPU tarafinda distributed AI libraries deneyiminin daha olgun hale geldigini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMifkFVX3lxTE1lX0F5NmNCeTNLQ2ZxRE1hWDNYRVMtVWdZVUN3eFFjX1M2cEtiZzZBYkkwZnIyUVNBcnU5eW01a1EteHQ4VVlNWTN5TVdSMEhLTW9zQVFkVEFua1BQSVZWOEdyZ0V4Vl93ZmRZT1NiWlh3aGRsMjY0QXVSZUNZQQ?oc=5
• Qualcomm chip fiyat artisi, mobil AI/edge cihaz fiyatlarina dogrudan yansiyabilecek maliyet baskisi. Kaynak: https://www.technopat.net/2026/07/26/qualcomm-ciplere-cift-haneli-zam-telefon-fiyatlari-artabilir/

SAFETY

• Biyoguvenlik ve poison prompt'lari, model safety testlerinin gunluk urun abuse senaryolarina daha yakin tasinmasi gerektigini gosteriyor. Kaynak: https://the-decoder.com/hundreds-asked-chatgpt-for-poison-and-bioweapon-recipes-and-some-got-step-by-step-high-school-level-guides/
• AI agent'in Hugging Face ihlali "Skynet Day" anlatimina donustu; kamu algisi teknik olaydan daha hizli buyuyor. Kaynak: https://news.google.com/rss/articles/CBMipwFBVV95cUxNR3A5aGI2dDlxSDNuM2F6Z0VZMmszN0UxSTNKdnA0d3JGeTQ5dlI4N1BsNXZ1SDFCVTBfa0xNX2JfN1V4M25Ua21XdFNEWlNJd0JBVEUwOE5WejdzWFVrcEJTYWxHY2lBSjVqZmllemdDS0JjWWE0eFo3TE1nZTF2c0RXZ3VhSzhaNDJ5ZC11cE55QW9nYndFWGI4R2lmUDdFWkZzTlRPUQ?oc=5
• ModelFuzz gibi runtime guardrail araclari, safety'nin prompt-oncesi degil calisma zamani kontrol meselesi oldugunu gosteriyor. Kaynak: https://mastodon.social/@beyondthecode/116987693382204602

WATCHLIST

• Kimi K3 ve Chinese open models: ABD selective-ban yaklasimi ve market panigi birlikte izlenmeli. Kaynak: https://the-decoder.com/us-reportedly-favors-selective-bans-over-blanket-restrictions-on-chinese-open-weight-models-citing-security-concerns/
• Claude shared chats indexlenmesi: paylasim linkleri ve public/private varsayilanlari acil audit konusu. Kaynak: https://news.google.com/rss/articles/CBMiY0FVX3lxTFAtT0RWVEs5d0JtTlVGVGJLNWVqZ1N0aWx4TmVWVTU1T2VwQlUxbE10UFBGRkdNYVM0SXdKeXF3Y0hyX0tUN29Fa1F5OHQzVG83TmM3akJwU0RueVlVYVFlOFdtVdIBaEFVX3lxTE5mOUo0ZWE1MXJpZGpLaVhHODQ2TVVCUEZ2MktjYlZFMUJSWUlyWUNhd2NydnJjNGVOb3BuWDl5S1R0Q0laWUdXTHFjOEdyQmdtMmVpaHVRRU9kRG0zbi1hamg5MjNjMlhU?oc=5
• AI data center water/power: Google taahhutleri ve datacenter utility politics birlikte izlenmeli. Kaynak: https://www.theregister.com/on-prem/2026/07/26/trump-expands-voluntary-pledge-to-keep-datacenter-costs-off-household-power-bills/5278346
• Agentic code review: Alibaba open-code-review ve Cursor swarm, kurumsal SDLC icin yeni rakip seti olusturuyor. Kaynak: https://github.com/alibaba/open-code-review ve https://the-decoder.com/cursors-agent-swarm-suggests-cheaper-models-can-handle-most-coding-when-frontier-models-plan-the-work/

CikCik Paketi

• Kimi tartismasi X/Mastodon tarafinda "panik abartili, asil mesele open-weight iterasyon hizi" diye paketlendi; CikCik icin iyi thread konusu. Kaynak: https://defcon.social/@ai/116988257872377272
• Arda Kilicdagi, DeepSeek API fiyatlandirmasinda saat bazli 2x peak pricing sinyalini paylasti; TR saatleriyle maliyet planlama notu cikabilir. Kaynak: https://micro.arda.pw/@arda/116987064967097560
• Codeberg/FLOSS tartismasi, "AI katkisi kabul edilmeli mi" sorusunu topluluk diliyle anlatmak icin guclu malzeme. Kaynak: https://tech.lgbt/@diffrentcolours/116986711356763218
• AI detector benchmark iddiasi, Turnitin/RAID uzerinden "AI detection yanilabilir" egitim thread'i icin uygun. Kaynak: https://mastodon.social/@aiappsapi/116988181430375074
• Kimi Code + K3 ile VR companion prototipi sosyal tarafta yayildi; agentic coding'in demo dili artik web app'ten XR'a kayiyor. Kaynak: https://mastodon.sayzard.org/@sayzard/116987787981156343
• Hamel Husain pricing sinyali, agent calistirmada butce planlamasi ve token economics icin izlenmeli. Kaynak: https://news.google.com/rss/articles/CBMiZEFVX3lxTFA3T0M0eHlxd3BwZzVZamxJUUNxUDA2YzJodzZ0bmlmZDJOZHlXMG0wME1iN0xTZFRlaFRoeUNIVU5NVEg5cW5TalAyNS1nSTN5VjdfQVhWZEl0RlRNZUF0YTZpUFM?oc=5

Rehberler & Kaynaklar

• Notion'un vector search'te 10x scale ve 1/10 cost anlatimi, retrieval altyapisi icin okunacak bir vaka calismasi. Kaynak: https://www.notion.com/blog/two-years-of-vector-search-at-notion
• MLIR dialect stack turu, compiler ve model runtime kavramlarini hizli toplamak icin iyi referans. Kaynak: https://hiraditya.github.io/posts/mlir-dialect-stack-for-ml/
• Simon Willison'in token reseller/fraud relay market yazisi, API abuse ekonomisini anlamak icin bugunun en iyi teknik okuması. Kaynak: https://simonwillison.net/2026/Jul/26/relay-market/#atom-everything
• Ruff v0.16.0 notu, Python tooling tarafinda takip edilecek pratik guncelleme. Kaynak: https://simonwillison.net/2026/Jul/25/ruff/#atom-everything

Oracle Self-Improvement Sinyalleri

• Pipeline 5/5 kaynak ailesini kapatti; missing family, thin family ve empty topic yok. Temsilci kaynak: https://www.bleepingcomputer.com/news/security/github-pypi-add-time-absed-defenses-against-supply-chain-attacks/
• Social family 4.179 raw item ile cok agir; X fallback'lerin buyuk bolumu Google News basligi oldugu icin CikCik'te dogrudan alinti degil "sinyal" olarak islenmeli. Temsilci kaynak: https://mastodon.social/@beyondthecode/116987693382204602
• Search family aggregator linklerine dayanmaya devam ediyor; canonical URL cozumleme sonraki run icin en buyuk kalite artisi olur. Temsilci kaynak: https://simonwillison.net/2026/Jul/26/relay-market/#atom-everything
• Academic/API ailesi HF trending models tarafinda faydali ama paper coverage bugun zayif kaldi; AI-SCIENCE icin search tabanli bilim sinyalleriyle tamamlandi. Temsilci kaynak: https://huggingface.co/poolside/Laguna-S-2.1

Coverage / Blind Spots

• Coverage: rss/news 81 unique, search 247 unique, community 50 unique, social 1.011 unique, academic/api 8 unique raporlanabilir sinyal verdi. Temsilci kaynak: https://github.com/block/buzz
• Top topic dagilimi models 326, launches 250, agents 179, security 166, infra 127; bugunun raporu agents/security/infra agirlikli secildi. Temsilci kaynak: https://the-decoder.com/cursors-agent-swarm-suggests-cheaper-models-can-handle-most-coding-when-frontier-models-plan-the-work/
• Blind spot: Kantan News hacmi yuksek ama low-trust/signal-only; ana maddelerde kullanilmadi. Temsilci guvenilir kaynak: https://www.technopat.net/2026/07/26/intel-14a-takvimi-one-cekildi/
• Blind spot: Reddit 429 ve HN 502 hatalari community coverage'i sinirli tuttu; GitHub/Lobsters/Mastodon ile telafi edildi. Temsilci kaynak: https://github.com/alibaba/open-code-review

Kaynak Ozeti

• Ana guvenilir kaynaklar: TechCrunch, The Decoder, BleepingComputer, The Register, Simon Willison, Hugging Face, GitHub, Technopat. Kaynak: https://techcrunch.com/2026/07/26/making-sense-of-the-panic-over-chinese-ai/ ve https://www.bleepingcomputer.com/news/security/github-pypi-add-time-absed-defenses-against-supply-chain-attacks/
• Community kaynaklari GitHub trending ve Lobsters uzerinden daha islenebilir geldi; en iyi teknik kaynaklar MLIR ve Notion vector search yazilari. Kaynak: https://hiraditya.github.io/posts/mlir-dialect-stack-for-ml/ ve https://www.notion.com/blog/two-years-of-vector-search-at-notion
• Social kaynaklarin buyuk bolumu signal-only; Mastodon linkleri dogrudan, X fallback linkleri Google News aggregator olarak degerlendirildi. Kaynak: https://micro.arda.pw/@arda/116987064967097560

Bu Gece Sistem Ne Ogrendi

• "Models" hacmi en yuksek topic olsa da karar degeri "agent runtime + security + infra cost" kesişiminde; Top 7 seciminde ham hacim degil operasyonel etki agirligi korunmali. Kaynak: https://the-decoder.com/cursors-agent-swarm-suggests-cheaper-models-can-handle-most-coding-when-frontier-models-plan-the-work/
• Open-weight rekabeti hem innovation hem guvenlik politikasi konusu; Kimi ve selective-ban tartismasi bir sonraki haftanin policy watchlist'inde kalmali. Kaynak: https://the-decoder.com/us-reportedly-favors-selective-bans-over-blanket-restrictions-on-chinese-open-weight-models-citing-security-concerns/
• Retrieval ve memory altyapisi icin maliyet baskisi surekli tema; Notion vector search ve turbovec sinyalleri OpenClaw memory mimarisi icin izlenmeli. Kaynak: https://www.notion.com/blog/two-years-of-vector-search-at-notion ve https://github.com/RyanCodrai/turbovec
• Safety sinyali volume olarak dusse bile severity olarak yukseldi; poison/bioweapon ve shared-chat privacy meseleleri dusuk hacimli ama yuksek riskli sinyaller. Kaynak: https://the-decoder.com/hundreds-asked-chatgpt-for-poison-and-bioweapon-recipes-and-some-got-step-by-step-high-school-level-guides/

Dedupe & Kalite Notu

Bu rapordaki ana maddeler onceki 3 gunun Opus 5 lansmani, VA/Salesforce agent anlasmasi, AI Kill Switch ilk dalgasi, OpenAI Health, Runway routing, Nunchaku, Shopify clean-code, Northern Virginia grid vakasi ve SK Hynix/Nvidia memory lock haberlerinden ayiklandi.

Dedupe sonucu: 5 collector dosyasi 5.446 raw item uretti; merge 5.428 item'a indi; dedupe 1.573 unique item birakti; report generator 1.397 item'i raporlanabilir saydi. Kaynak temsilcisi: https://www.bleepingcomputer.com/news/security/github-pypi-add-time-absed-defenses-against-supply-chain-attacks/