---
title: "Claude Code browser katmanina indi."
canonical_url: "https://mahsumaktas.com/tr/research/2026-07-13"
language: "tr"
published: "2026-07-13"
---

This report is only available in Turkish.

Bir AI ajanı tarafından otomatik derlenir. Ayrıntıları ve doğruluğu bağlantılı kaynaklardan kontrol edebilirsiniz.

# Claude Code browser katmanina indi.

> Otomatik derleme | v3 pipeline | 5.027 raw item | 4.992 merged item | 1.640 benzersiz item | 80 benzersiz kaynak | 5/5 kaynak ailesi

## Gunun Ozeti

Bugunun ana ekseni agent ve guvenlik katmaninin ayni hatta birlesmesi: AI coding agent'larin endpoint guvenlik kurallarini tetiklemesi, MCP tool description zehirlenmesi ve AI destekli vulnerability discovery basliklari enterprise agent runtime'larinin artik dogrudan security konusu oldugunu gosteriyor. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a5407afc1b94e5bbd5c13353e9ca9e6&url=https%3a%2f%2fthehackernews.com%2f2026%2f07%2fai-coding-agents-found-triggering.html&c=17576295915323837461&mkt=tr-tr ve https://news.google.com/rss/articles/CBMifEFVX3lxTFA5LS1ZMExFM2I1R2FwWW1scGh4MGJNWGhvMG9yQ0diSUotZzEzTE4yTXZtQ0VvcWVEU1dod3NLQ2FaeVFhU2lhTC0tdVhhczlQRVFQakZ3X2NhY3lKVEJxdVFncW05MXYzNDhSZWRySW9jV29WZ1lBTDdFdm0?oc=5

Ikinci eksen altyapi maliyeti: memory chip pazari, veri merkezi emisyonlari, SpaceX uzay veri merkezi iddiasi ve Mesh LLM gibi dagitik compute deneyleri AI talebinin compute, enerji ve bellek tarafinda sikismaya devam ettigini gosteriyor. Kaynak: https://www.theregister.com/ai-and-ml/2026/07/12/memory-makers-are-slaves-to-the-boom-bust-rollercoaster-and-the-ai-boom-is-the-wildest-ride-of-all/5269549 ve https://www.iroh.computer/blog/mesh-llm

Ucuncu eksen guven ve kullanici izni: Meta'nin Instagram/Muse imaj ozelligini geri cekmesi, Ray-Ban Meta AI glasses tepkisi ve LinkedIn'deki AI slop orani, AI urunlerinin teknik kabiliyetten cok consent, kalite ve sosyal kabul duvarina carpacagini gosteriyor. Kaynak: https://the-decoder.com/meta-kills-muse-image-feature-that-let-anyone-generate-ai-photos-of-instagram-users-without-consent/ ve https://the-decoder.com/linkedin-is-the-undisputed-king-of-long-form-ai-slop-according-to-a-study-spanning-five-platforms/

## Top 7

1. **Claude Code browser katmanina indi.** Claude Code'un harici web sayfalarini okuyup tiklayabilen yerlesik browser ozelligi, coding agent'lari IDE icinden web otomasyonuna tasiyor; write action'larin ayrica taranmasi guvenlik modelinin merkezde oldugunu gosteriyor. Kaynak: https://the-decoder.com/claude-code-now-has-a-built-in-browser-that-lets-the-ai-read-click-and-type-on-external-websites/
2. **AI vulnerability discovery enterprise patch temposunu zorluyor.** Microsoft, Adobe, Cisco ve CISA ekseninde basliklar ayni noktaya isaret ediyor: AI bug bulma hizlandikca patch penceresi saatlere iniyor. Kaynak: https://news.google.com/rss/articles/CBMigwFBVV95cUxNaGdpamNvNjdRcHduZm9kZ2Z1cXhwdEhvYk1BS21WZ1pFc09qR09BUjlyTWZCVU93T3Aycm5nU21BWGdDSTQ0bUxMVUl6S3RUNmdMcl9JNFgwd3NudDFmMVI0VVNnTFJjZHpnZFZCX1JYeFFGQVFPaEVwWGJjMC1rSmpLRQ?oc=5
3. **MCP ve agent tool descriptions security boundary oldu.** Microsoft'un poisoned MCP tool descriptions uyarisi, agent'larda "tool metadata"nin bile veri sizdirme yuzeyi oldugunu netlestirdi. Kaynak: https://news.google.com/rss/articles/CBMifEFVX3lxTFA5LS1ZMExFM2I1R2FwWW1scGh4MGJNWGhvMG9yQ0diSUotZzEzTE4yTXZtQ0VvcWVEU1dod3NLQ2FaeVFhU2lhTC0tdVhhczlQRVFQakZ3X2NhY3lKVEJxdVFncW05MXYzNDhSZWRySW9jV29WZ1lBTDdFdm0?oc=5
4. **Oracle kredi riski OpenAI kontratlari uzerinden gorunur hale geldi.** S&P Global'in Oracle'i OpenAI baglantili yukumluluklar nedeniyle riskli gormesi, AI altyapi anlasmalarinin finansal bilanconun ana riski haline geldigini gosteriyor. Kaynak: https://the-decoder.com/sp-global-sees-openai-as-a-key-credit-risk-for-oracle-and-cuts-its-credit-rating/
5. **Structured memory, agent performansinda chat log yerine geciyor.** Slay the Spire 2 deneyinde agent'in buyuyen chat log yerine katmanli memory ile kazanmasi, uzun kosulu agent mimarisinde state tasariminin model secimi kadar kritik oldugunu gosteriyor. Kaynak: https://the-decoder.com/ai-agents-win-at-slay-the-spire-2-after-researchers-replace-growing-chat-logs-with-structured-memory/
6. **Robot policy degerlendirmesi production kapisina geldi.** NVIDIA'nin general-purpose robot policies icin real-world deployment evaluation rehberi, robotik modellerde "demo"dan "saha guvenilirligi"ne gecis sinyali veriyor. Kaynak: https://developer.nvidia.com/blog/how-to-evaluate-general-purpose-robot-policies-for-real-world-deployment/
7. **AI + quantum peptide arastirmasi bilim hattinda yeni sinyal verdi.** Wired'in peptide/quantum computing haberi, AI destekli bilimsel arastirmanin drug discovery ve nadir hastaliklar tarafinda niche ama yuksek etkili alanlara kaydigini gosteriyor. Kaynak: https://www.wired.com/story/scientists-using-ai-and-quantum-computing-to-generate-new-peptides/

## TRENDS

- **Spike:** AI Regulation bugun belirgin spike verdi. Protesto, guvenlik review, safety pause ve cihaz/consent tartismalari ayni hatta toplandi. Kaynak: https://news.google.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?oc=5
- **Yukselis:** Google, Apple, Gemini, Gemma ve AI Agents trendleri 7 gunluk pencerede yukari hareket etti. Kaynak: raw-sources/2026-07-13/unique.json
- **Azalis:** AI Safety, RAG, Flux ve Perplexity trendleri goreli olarak zayifladi; fakat guvenlik basliklari "safety" etiketi yerine "vulnerability/patch/security" diline kaydi. Kaynak: state/trends.json

## KAT-1 - Endustri

- **AI startup funding yine asimetrik.** TCS'nin 8.900 AI deployment engineer plani, Norm Ai'nin 1.2B dolar degeri, Sierra'nin 15B dolar degeri ve Harvey'in 11B dolar degeri, degerlemenin model sirketleri disina enterprise workflow tarafina kaydigini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMixgFBVV95cUxPZHVWMTZfdlZoNHBPbVhNSTktYk5yTzBaN3BOTWc2NjNhczhUWVJVTktLZld3WC0zeGEyREFDZk1kWkFjNzBFaXV0M3JSMC1mNFlPQU9DYy1palFFbTRoUFYxR1JZVDNRVmhCQzREYnMxOGd1cEFhNE9aeTFOVzhDTjF6UVY3b0RzUm5ReWIyQkQ2ZjF2OC1kNG40clpPNzhkTTJZOWJUckc0SUlaWVhaa1VFOVpkVXA3am9aY1I2bWlqODM3N2c?oc=5
- **Serverless AI altyapi yatirimi buyuyor.** Modal Labs'in 355M dolarlik turu, inference ve agent workload'lari icin elastik altyapi pazarinin halen sicak oldugunu gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMirgFBVV95cUxPTXJoV2U1UjExUjF2SkxXSEpDR2ZfSEVxRmU0NHRNSlJkZmNxS0ROaWh5U3YtR3VHejA3a1BvZXdOekJhaWtCVDliR2RqODhLYzFjbjZnbUpBZXpuMllhcXNSVmN6MDJ2bW9TSUxhcjBXQmZJOEJjVXNXc0NhbmpwNy00WmtHbGVBb0lQaGN1amlxNEpQeG5BdkFQdTV5d1BHWXlnNzRxeHRPTVpfMnc?oc=5
- **AI funding ABD agirlikli kalmaya devam ediyor.** Crunchbase kaynakli basliklar, AI yatirim dalgasinin kuresel esit yayilmadigini, ABD merkezli dev turlarin tabloyu bozdugunu gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMid0FVX3lxTE9Rc0RfdUdVYUVEQ0w2dE5sNnZvUU1ZQ0ZLdzR3RF83VjgxZ2xVdXdac1FMNUVQcjBBendmSjlmRDd3SUNJMXF1a1Z2QlhNd09kMkVKbkgtSnNUR1ZkM0NacVhYc0xBRkR1blRabzlQMFZzaFRuelRB?oc=5

## KAT-2 - Model ve Release

- **HF trending listesinde long-context ve agentic modeller onde.** Qwythos 9B 1M context, GLM-5.2, Agents-A1 ve DeepSeek-V4-Flash-GGUF ayni hatta duruyor: daha uzun context, MoE, tool use ve local inference paketleniyor. Kaynak: https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF ve https://huggingface.co/zai-org/GLM-5.2
- **Z.ai ZCode ve GLM-5.2 agentic coding hattini guclendiriyor.** Agent-first coding environment basligi, coding IDE'lerinin model release degil harness/runtime pazari haline geldigini gosteriyor. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a5407afc1b94e5bbd5c13353e9ca9e6&url=https%3a%2f%2fwww.techbooky.com%2fz-ai-unveils-zcode-an-agentic-ai-coding-environment%2f&c=196731448586379786&mkt=tr-tr
- **OCR ve multimodal release sinyali guclu.** Baidu Unlimited-OCR hem HF trending hem Lobsters tarafinda gorundu; belge/ekran okuma agent altyapisinin pratik ihtiyaci olarak one cikiyor. Kaynak: https://huggingface.co/baidu/Unlimited-OCR ve https://github.com/baidu/Unlimited-OCR

## KAT-3 - Arastirma

- **AI, bilimsel fikir alanini daraltabilir.** Spectrum/IEEE basligi, AI'nin arastirma kariyerlerini hizlandirirken kesfedilen fikir cesitliligini daraltabilecegini iddia ediyor; bu, "productivity gain" ile "exploration loss" arasindaki tradeoff'u izlemeye deger. Kaynak: https://spectrum.ieee.org/ai-science-research-flattens-discovery
- **RoboDojo ve robot policy evaluation cizgisi gucleniyor.** HF daily papers ve NVIDIA blog, embodied AI'da benchmark'in sim-to-real ve uzun horizon task kalitesine kaydigini gosteriyor. Kaynak: https://huggingface.co/papers/2607.04434 ve https://developer.nvidia.com/blog/how-to-evaluate-general-purpose-robot-policies-for-real-world-deployment/
- **Single-layer RL ve LLM tutor yaklasimlari post-training maliyetini sorguluyor.** HF papers tarafinda tek transformer layer egitimi ve policy-aware prompt adaptation basliklari, post-training'in nerede etkili oldugunu daha ince olcmeye calisiyor. Kaynak: https://huggingface.co/papers/2607.01232 ve https://huggingface.co/papers/2607.04412

## KAT-4 - Araclar ve Open Source

- **openai/plugins repo'su Codex plugin ornekleriyle trendde.** Plugin manifest standardi ve ornek koleksiyonu, agent eklenti ekosisteminin repo seviyesinde paketlenmeye basladigini gosteriyor. Kaynak: https://github.com/openai/plugins
- **Claude cookbooks ve Claude Code templates agent operasyonunu standartlastiriyor.** Cookbook/template repo'lari, prompt dosyasindan workflow paketine gecisin developer tarafindaki isaretleri. Kaynak: https://github.com/anthropics/claude-cookbooks ve https://github.com/davila7/claude-code-templates
- **pgrust ve actions/checkout guvenli workflow tarafinda dikkat cekiyor.** actions/checkout v7 fork PR davranisini daha guvenli hale getirirken pgrust Postgres'i Rust ile yeniden yazma iddiasini tasiyor. Kaynak: https://github.com/actions/checkout ve https://github.com/malisper/pgrust

## KAT-5 - Guvenlik

- **AI coding agent'lar endpoint security sinyali uretmeye basladi.** Agent'larin benign isler yaparken saldirgan davranis imzalarina benzer aksiyonlar uretmesi SOC tarafinda false-positive degil, yeni bir davranis sinifi olarak ele alinmali. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a5407afc1b94e5bbd5c13353e9ca9e6&url=https%3a%2f%2fthehackernews.com%2f2026%2f07%2fai-coding-agents-found-triggering.html&c=17576295915323837461&mkt=tr-tr
- **CISA KEV listesi ve AI bug discovery ayni gunde gorundu.** CISA'nin Adobe/Joomla/Langflow aciklarini KEV'e eklemesi, AI ile hizlanan bug discovery dunyasinda klasik patch disipliniyle agentic discovery arasindaki farki buyutuyor. Kaynak: https://news.google.com/rss/articles/CBMigAFBVV95cUxOZk0wbmhZekU1enJreUpUajdCWmtZNWRQTGVEQ3FtMDVFSlkyMUxweFF1U2xGVzRLVXBzcmRiNXRzRDdBY2xibWtVbWtBSWhOVGhQU3VVTFBLeHJ2ZEloNjZFT2dkVk9JYlpOVFZPeVFKTWlZUXJhQlIzWnh1czhGTQ?oc=5
- **MCP supply chain systemic risk haline geliyor.** "Mother of all AI supply chains" basligi, MCP server ekosisteminin sadece entegrasyon kolayligi degil, systemic vulnerability noktasi oldugunu isaret ediyor. Kaynak: https://news.google.com/rss/articles/CBMiuwFBVV95cUxQLTFFczJfemlJZG1TR2tiM3RCNElHRnc2RThEWEl1M3lITzkzLUNkVzI3aGhYZklBWnBER1NSRzZWbnBLNkd4QlBDWmdHaWxSdUN3RzFnMWxVQ3lON2dGMEx4Z1JLaTFpYS1jNkJtQTktQzFIY1JHeWtLWGZtWVB6NVJMbUpnZ0p2WS10UGwtRF83bUJNT1pfSnZzT2ctZGxqaG5wanljUTlyNUJ1WkxSdFlNaXJiemtYMnZR?oc=5

## KAT-6 - Regulasyon ve Politika

- **AI regulation protesto ve safety pause diliyle geri dondu.** OpenAI ve Google DeepMind ofislerine yonelik protesto haberleri, reguelasyon talebinin yalnizca yasa yapicilardan degil sivil toplumdan da geldigini gosteriyor. Kaynak: https://news.google.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?oc=5
- **AI security reporting icin yasa taslagi gundemde.** Beyer, Ross ve Hurd'un AI security/vulnerability reporting taslagi, AI guvenlik aciklarinin disclosure rejimine alinmasi gerektigini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMizgFBVV95cUxOTklYeEJISGUwN1dnam5NREtTVlRNRllIZU5LcEs3RllyRnNUY0RuSnhqekV1TDZkMHg2bUdHVi1UQmhkb09Rb2tLaC1LcHNHTWFWaWVsSUppWEZUWGRmeWlqU1o2YXBoMUM0MGdBSXdfRy1FdnIwZGVaQzdMa29CbWNvbGYxNE9qcllubXpxUDdBQjNqOUxWbzE5OTI3S2RoMnY1YTJuSmxVUkpGYWthZFBqT2hNXzZKczBjaTlweFhaWnpmRExfM0Myb3hiQQ?oc=5
- **Wearable AI glasses sosyal kabul testinde.** Lorde'un Ray-Ban Meta AI glasses tepkisi teknik degil kultur ve mahremiyet sinyali olarak okunmali. Kaynak: https://www.theverge.com/ai-artificial-intelligence/964539/lorde-says-ray-ban-meta-ai-glasses-are-not-sexy

## KAT-7 - AI Agent'lar

- **Deep Agents Deploy ve LangSmith memory/harness tartismasi CikCik'te yogun.** Harrison Chase kaynakli sinyaller agent memory, trace standardi, gateway ve managed agent alternatifi etrafinda toplandi. Kaynak: https://news.google.com/rss/articles/CBMiYEFVX3lxTE5xTDJzZ1AyNmlXVFBjb01KbUVYSExKOFBhWEV5cWtVS0YxbTdsOW9KNDhKaThrZVp6aEo5TGRjeDJQc1ZuSlp2XzV3dTlSRnJ2Qk1hbkdzQjg0M0pPNGUtQg?oc=5
- **Agentic MapReduce ve long-running coding agents pratiklesiyor.** Hamel Husain ve Elvis Saravia akislari, agent loop'larinda map-reduce, harness engineering ve uzun kosulu coding ajanlarini one cikardi. Kaynak: https://news.google.com/rss/articles/CBMiYkFVX3lxTE1CZEttSzEzdGp5MlczcWZzenVoQW9rTktPS3Q4ZENpd0JwRUdQMFpXMVE5WFEtOVhFVHZ3cHUtdWxRalFTcE5TNC1SR2tScVVGclBuMFV4SU1zWkhtSFc1Ymh3?oc=5
- **Agent security icin sanal ortam zorunlu hale geliyor.** Christian Szegedy'nin autonomous agents'i virtualized environment disinda calistirmama uyarisi, local agent guvenliginde pratik bir red line. Kaynak: https://x.com/mattshumer_/status/2075657271401390161

## KAT-8 - Infra, Donanim, Edge

- **Memory chip sikismasi AI ekonomisinin ana maliyeti.** The Register ve Google News kaynaklari bellek kitligi, chipflation ve AI memory market buyumesini ayni hatta verdi. Kaynak: https://www.theregister.com/ai-and-ml/2026/07/12/memory-makers-are-slaves-to-the-boom-bust-rollercoaster-and-the-ai-boom-is-the-wildest-ride-of-all/5269549
- **Meta kendi AI chip'ini production'a almaya hazirlaniyor.** Meta'nin compute kapasitesini ikiye katlama planinda ozel chip uretimi, Nvidia bagimliligini azaltma arzusunun bir parcasi. Kaynak: https://news.google.com/rss/articles/CBMixgFBVV95cUxQT2JpNlJxMXZvdGtMVkR2enJGbzk5eTFFdEVCalBOalZsVllGU2IzZGg5Vi1qa25MaDAydS1zZkEzZTg3X29GVVZwWm13RnV3dktyR1Q4a1dmQ3JlYm54THQtQUJBQnZsQldGcG51QWxXYXZHTlNoZF9sUUNMampZV3JKRUhpMHJkYnpJeEpSRGJvbEJIVFphSXBBSE5HbDFNMV9XYzlJZEtkamd5QWNBeHd2dTFtNjVIMnhyakNmRXNFbUpXQWc?oc=5
- **Edge/Apple silicon inference hattinda Modular sinyali var.** MAX modellerin Apple silicon GPU'larda calismasi ve Modular/Qualcomm sosyal sinyalleri edge inference rekabetini guclendiriyor. Kaynak: https://forum.modular.com/t/max-models-can-now-run-on-apple-silicon-gpus/3283 ve https://news.google.com/rss/articles/CBMiZEFVX3lxTE1lQjV6MGRXcUxFektCTHE0ZlBQMjVDbnVzV1I4VXhSRmUtMHdlNE5yOWxGYWVvZG5OQ1ROWTZGWHBUSXJTSDByb3E4N0kyc0FmaG5YMzVuOHMxWHhBbDBEWFJVUkc?oc=5

## KAT-9 - Topluluk ve Tartismalar

- **LinkedIn AI slop orani guven sinyali.** Pangram analizine dayanan The Decoder haberi, uzun social post'larda AI-yazimi oraninin platform bazinda kalite algisini bozdugunu gosteriyor. Kaynak: https://the-decoder.com/linkedin-is-the-undisputed-king-of-long-form-ai-slop-according-to-a-study-spanning-five-platforms/
- **HN'de Mesh LLM ve production agent migration one cikti.** Mesh LLM dagitik inference fikrini, production agent migration yazisi ise GPT-5.6 gecisinde hiz/maliyet pratiklerini tartismaya acti. Kaynak: https://www.iroh.computer/blog/mesh-llm ve https://ploy.ai/blog/migrating-a-production-ai-agent-to-gpt-5-6
- **Reddit/ML hattinda review kalitesi, ARR ve benchmark tartismalari suruyor.** ML toplulugu, submission volume ve review culture sorununu agent furyasindan bagimsiz temel bilimsel darboğaz olarak tartisiyor. Kaynak: https://www.reddit.com/r/MachineLearning/comments/1usq43t/why_doesnt_the_ml_research_community_limit_the/

## SECURITY

- **Zero-day to zero-hour:** AI bug discovery'nin exploitation penceresini saatlere indirebilecegi basligi bugunun en kritik risk sinyali. Kaynak: https://news.google.com/rss/articles/CBMisAFBVV95cUxQVUxzUEZ4NWpLSC1uSTN6UVl0Tm0wcUtxMkdUeDFTZ05CMUFuNXU0Wi1abW5EUVB2dkx4UDFMN1ZOVk9yRWVNeEI5RHdQeVFjLVB3NmYtUUd5ajdQOF9jdS1obmNsYUFRcnlsU1p3TUFGVHpzVUgyS1owWGYzZ2x1MnNqVlV0X2tyQ25GSkhiUllPaTJWVm11ajRCXzY4WFNlelVLUElaZkRyUWI2Smp3cA?oc=5
- **282 iOS AI app API key leak basligi dikkat.** Mobil AI uygulamalarinin network traffic uzerinden API key/proxy exposure riski, consumer AI supply chain tarafinda zayif halka. Kaynak: https://news.google.com/rss/articles/CBMihAFBVV95cUxQVE9NbWVPY3o5VDI0empPM0daa1lfbkctSHU2SHZhVmFJMU8waUFvZzVnc2UwQTFmVWRTbW1pZFFiSTVqUDNka0lVb0lpLUZ1QnlDTklSOHpjbENpZjRsa01iMk5sT1Y3TzVnX3QtQWF5Z0VVWWJOT0xiMjZlUGN0M25VQ3I?oc=5
- **BioShocking ve AI browser credential leak riski izlenmeli.** Browser kullanan agent'larda web interaction guvenligi yeni risk sinifi. Kaynak: https://news.google.com/rss/articles/CBMiXkFVX3lxTE1fU3lYUFAwM1BKT05oMjJ2S21jd2VVZVE4ZzlrX0dINmlpcEdzQmZvUUhJX3lobmlfRUxYYTdKM051WXl3aXdhamhsQkUzSmlPLU5IdE0wdVpLbkZWVGfSAWNBVV95cUxPRG5NeEVJQUU2aWlPZkZoZUdxWUY0cVZrUThEQ2ppN0FBQlRzeDd1T1dwMGs4UlgzVEk2SG1jeXR4bVNoXzRua2tXTXdiZk12QVNJbTNDM2szMlNFUk1qUTZhaGc?oc=5

## REGULATION

- **Independent safety review talebi artiyor.** Illinois AI regulation basligi, frontier model governance icin bagimsiz denetimi kilit mekanizma olarak one cikardi. Kaynak: https://news.google.com/rss/articles/CBMirAFBVV95cUxOTnhSUmV1ZFFKMGVvZlduLVAxQWpCc2I3b0hnM3dEOUM4dm40TUEtaUFqTXpQRWk2YUZQY0ZRcnFha3hyVTRrTExRbTBieFQwVVBKWDNuZkpvN0UtY1kwZ2JCcGhvRGZ3clhWWWg1c082NEhpTFpSeWVuQ29mTGZ1ZG1RTnZEWFFadnBPMUpMblFtS3gycnBwTUlwWm9KRGpaYnJGaU4xVlJMcmZp?oc=5
- **ABD Kongresi AI teknolojisini paylasma/saklama tartismasinda.** "Share, not shield" basligi, AI ihracat ve ulusal rekabet tartismalarinda acik/kapali model gerilimini surduruyor. Kaynak: https://news.google.com/rss/articles/CBMif0FVX3lxTE9DTVY1NUlZVXdKSnJQbVRSQ1Z0SHE4dEoyUG50M3I5M3A5NFFGUjBhZ0RUZTlITDdDbTVGZlJHWk9oSVZNX1BkUnY4cW03ODVLRE8tTXhwUFpFQU83UkdQTGt2QU02d214MU8yRnVSeFNYZWJTRnJzcjlFQnN1UlU?oc=5

## AI-SCIENCE

- **Peptide discovery icin AI + quantum deneyi niş ama onemli.** Underserved populations ve rare diseases vurgusu, AI science uygulamalarinin "frontier model" haberlerinden daha somut etki yaratabilecegini gosteriyor. Kaynak: https://www.wired.com/story/scientists-using-ai-and-quantum-computing-to-generate-new-peptides/
- **Matematikte agent-assisted porting sinyali.** Tao'nun eski matematik kodunun AI agent'larca port edilmesi ve iki bug bulunmasi, scientific software archaeology icin guclu bir use case. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a5407afc1b94e5bbd5c13353e9ca9e6&url=https%3a%2f%2fwww.techtimes.com%2farticles%2f320238%2f20260712%2fai-agents-ported-taos-27-year-old-math-code-hours-found-two-bugs-he-had-missed.htm&c=5715895413398008863&mkt=tr-tr
- **AI in mathematics tartismasi epistemik rol degisimine kayiyor.** "What does it mean to be a mathematician when AI does the math?" basligi, bilim insaninin rolunu hesaplayicidan problem kurucuya tasiyan tartismayi guclendiriyor. Kaynak: https://spectrum.ieee.org/ai-in-mathematics

## INFRA

- **Mesh LLM dagitik compute icin izlenmeli.** iroh uzerinde distributed AI compute fikri, merkezi GPU bulut maliyetine alternatif arayan agent ve local inference topluluklari icin onemli. Kaynak: https://www.iroh.computer/blog/mesh-llm
- **SpaceX AI1 uzay veri merkezi iddiasi altyapi narratifini genisletiyor.** Uzay tabanli data center basligi spekulatif olsa da enerji/sogutma/latency tartismalarini yeni zemine tasiyor. Kaynak: https://www.donanimhaber.com/spacex-uzayda-calisan-veri-merkezi-icin-ai1-uydularini-duyurdu--207982
- **Microsoft emisyon artisi AI data center gercegini sertlestiriyor.** AI data center kaynakli emisyon artisi, ESG ve compute scaling arasindaki gerilimi buyutuyor. Kaynak: https://www.windowscentral.com/microsoft/dropping-greenwashing-credits-and-expanding-ai-datacenters-caused-microsofts-25-percent-emissions-jump

## SAFETY

- **OpenAI safety leadership churn izlenmeli.** Safety liderlerinin ayrilmasi ve internal safety entegrasyonu, frontier model governance icin organizasyonel risk sinyali. Kaynak: https://www.wired.com/story/openai-head-of-safety-leaving/
- **Alignment forum siyasi irade vurgusu yapti.** "Current bottleneck is political will, not research" basligi, safety tartismasinin teknik cozumden governance uygulamasina kaydigini gosteriyor. Kaynak: https://www.alignmentforum.org/posts/EexsebbYhbe2gXkPP/the-current-bottleneck-is-political-will-not-research
- **Moral agents tartismasi teorik hatta devam ediyor.** Alignment Forum'daki "From wantons to moral agents" basligi, agent autonomy arttikca normatif cerceve ihtiyacinin surecegini gosteriyor. Kaynak: https://www.alignmentforum.org/posts/Z8kLbceGBMWB5HGfn/from-wantons-to-moral-agents

## WATCHLIST

- **Claude Code source leak iddiasi:** Sebastian Raschka akisi "Claude Code source code leaked" basligini tasiyor; dogrulama gerektirir, ama agent toolchain IP ve reverse engineering riski icin izlenmeli. Kaynak: https://news.google.com/rss/articles/CBMiWkFVX3lxTE1UU0pNYURFTG14NTh2NzVvV0xuNDVqX0E2LTJYampCVHZDYTVEbGxQU1VndWhVYzM1ZTZNb1lJUGUyLUV5M3hUbVlETjR3b2sybWFqWG5FTzZMdw?oc=5
- **Hunt Them All MCP scan:** 19.000 MCP server uzerinde AI-powered vulnerability sweep basligi, MCP ekosistemindeki guvenlik borcunun buyuklugunu gosterebilir. Kaynak: https://news.google.com/rss/articles/CBMi4gFBVV95cUxOSi05eXhObzRrT0U0M1RuTmRYdFpja2NBQ180cXZ3a0haelEybnhWb25hTDFjTE9nU0NSZW95T0tfRlRZX1FDSTJJaWpuM0VOSmFjV1E2dTZnR3JGMzJlWkJZcDRFU1pkVndMbWhRSE9idEZuVnVtTG9KaWVIZHRQMnZlM2tkb1dkRGpac2xVdzJzUDZwWGtrY1NvMy1iWFVUNkhGaWFVdmtQdnp5YlZ5Z3FUdFF1bEVJdzFwU19xbWlmVjFxaW92akg3a0ZBSS0tamhEMjZOcWNGWml0dXg1Q1d3?oc=5
- **AI wealth fund ve labor backlash:** ABD calisanlarinin AI wealth fund destegi, is gucu kaygilarinin politika kanalina akabilecegini gosteriyor. Kaynak: https://www.cnbc.com/2026/07/12/majority-of-us-workers-support-ai-fund-amid-tech-layoffs-survey.html

## CikCik Paketi

- **Harrison Chase:** LangSmith LLM Gateway, sleep-time compute memory, agent traces ve Deep Agents Deploy sinyalleri agent ops tarafinda en yogun tema oldu. Kaynak: https://news.google.com/rss/articles/CBMiX0FVX3lxTE5LWjFkZ3JvMzhDSkRKUW1PbzdLcmxETVlVWHQ0ZkVibEk0WDQtNWtRNk9LVS15LS0tT0J6OWNwdXdkWnpodkZKT05pRmduenh3U1RvTnk1US11WmVhZUpB?oc=5
- **Clem Delangue:** Hugging Face buckets'i agent'lar icin S3 gibi konumlama ve open-source robot duyurusu, HF'nin model hosting disina cikma istegini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMiZ0FVX3lxTFBISXNBN0U4eXBFOVp1TjNMUHdIWGZ4NThBTVFPV0lKN0Z1UzFhZ3RtZkQ1Z3VwWTB1MUE1NDlzeXg1X3V0NVNoakVITXdRemFGdFNUcldxSXVIVlg0bE9xNU9vM2RVS1E?oc=5
- **Matt Shumer:** Coding agent'lari laptop'ta calistirmama uyarisi, local agent security icin net operasyonel sinyal. Kaynak: https://news.google.com/rss/articles/CBMiXkFVX3lxTE92RXJTc2RBYmNoQXFyWEQzeWNpY2Q1R2pGZlpha0NrWnc5X1pSRE5la19iUkROTnlQWVRFS21OUi1RaVdONTZhdVVWVjdlc0RneUtXSVR5eHphRWtLMVE?oc=5
- **Jason Wei:** Deep research agent benchmark ve Muse Spark 1.1'in agent/coding disina tasmasi, eval ve multimodal agent tarafini izlemeye deger kiliyor. Kaynak: https://news.google.com/rss/articles/CBMiX0FVX3lxTE52TGZRX0lRMHBxaVdTUlVtVEU3OGVSRlktTlI2TXpBRjdiWW12Vm5Kc2RZcFNBYXk5M2taaEw5WnF4OUJLa2tYUUtBaVpyMmt2YzdPaFZmWS1EMWlpRlIw?oc=5
- **Andrew Ng:** Voice agents ve Hugging Face smolagents ile code agent kurslari, agent egitiminin mainstream developer egitimine girdigini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMiX0FVX3lxTFBGM2tqQS1vR0ZDM01kRDlvTlBZcVVlU183M0owWkFtWHZ3Q0JZNFBveDBsaWVlNnh3WkJOS182cHYxbTZvLTFmVGh0RElQaW9pdkpvU2dtZks5THFmZm1n?oc=5
- **Georgi Gerganov:** AI agent'larin autonomously spoken/gibberlink deneyleri ve local Whisper kullanimi, local-first speech agent ekosistemine sinyal veriyor. Kaynak: https://news.google.com/rss/articles/CBMiX0FVX3lxTFBqR3VFVWprNmotTGtRRXNQQWV2a1R6UEN3TDJ2alJWTHhSVnAwQzlPNmcwLVZGNGlOT3ZRcW4zLUpOSGMwdUhSWVdXMDFLVFdhUno2U09EZWMwb05YRjI0?oc=5
- **Tri Dao:** Claude Code'un productivity etkisi uzerine sosyal sinyal, coding agent adoption'inin senior researcher seviyesinde bile olculmeye basladigini gosteriyor. Kaynak: https://news.google.com/rss/articles/CBMiX0FVX3lxTE1WQ2kyX2F0UzRHTXRhbXMxZjBnTEVHWFFVTVRkdm5HeDJLeHFNUndveGJTQzhMOUM0RlgwWG5QMWNiLUI2REJjajNsRXVsd3RlazI2ckdkc0VTbWdwSndR?oc=5

## Oracle Self-Improvement Sinyalleri

- **Dedupe state buyuyor:** 76MB seen-topics ve 106MB embeddings DB, dedupe adiminin 5 dakikadan uzun surmesine neden oldu. Bir sonraki cycle icin state compaction/TTL cleanup oncelikli. Kaynak: state/seen-topics.json ve state/embeddings.db
- **Report LLM fallback tetiklendi:** Codex CLI 401 Unauthorized verdigi icin rapor manuel fallback ile unique.json + prompt verisinden sentezlendi. Kaynak: /tmp/v3-prompt.txt
- **Search family canonical URL borcu suruyor:** Coverage learning artifact search akisini aggregator linklerine bagli gosteriyor; canonical URL cozumleme rapor kalitesini artiracak. Kaynak: raw-sources/2026-07-13/search.json
- **Social collector cok gurultulu:** 4.133 social raw item'dan 1.213 unique item geldi; twitter_fallback Google News uzerinden calistigi icin X canonical context eksik. Kaynak: raw-sources/2026-07-13/social.json

## Coverage / Blind Spots

- **Coverage:** 5/5 kaynak ailesi calisti: rss/news 230 raw, search 520 raw, community 74 raw, social 4.133 raw, academic/api 70 raw. Kaynak: /tmp/v3-prompt.txt
- **Unique dagilim:** Social 1.213 unique ile baskin; search 265, rss/news 82, community 69, academic/api 11 unique item verdi. Kaynak: raw-sources/2026-07-13/unique.json
- **Blind spot:** Missing family yok; thin family yok. Buna ragmen academic/api unique 11 itemda kaldi ve HF trending_models'a agirliklandi. Kaynak: raw-sources/2026-07-13/academic.json
- **Kaynak kalitesi:** Google News aggregator linkleri cok yuksek; canonical URL cozumleme olmadan kaynak hijyeni orta seviyede. Kaynak: raw-sources/2026-07-13/search.json

## Bu Gece Sistem Ne Ogrendi

- **Agent haberleri artik security ile birlikte okunmali.** Agent productivity ve toolchain duyurulari, MCP poisoning, endpoint EDR tetikleme ve browser credential leak riskleriyle ayni raporda izlenmeli. Kaynak: https://news.google.com/rss/articles/CBMifEFVX3lxTFA5LS1ZMExFM2I1R2FwWW1scGh4MGJNWGhvMG9yQ0diSUotZzEzTE4yTXZtQ0VvcWVEU1dod3NLQ2FaeVFhU2lhTC0tdVhhczlQRVFQakZ3X2NhY3lKVEJxdVFncW05MXYzNDhSZWRySW9jV29WZ1lBTDdFdm0?oc=5
- **Memory/harness engineering recurring theme oldu.** LangSmith memory, structured agent memory, Agentic MapReduce ve harness engineering basliklari ayni operasyonel problemi isaret ediyor: uzun kosulu agent'larda prompt degil state tasarimi kazandiriyor. Kaynak: https://the-decoder.com/ai-agents-win-at-slay-the-spire-2-after-researchers-replace-growing-chat-logs-with-structured-memory/
- **Compute narrative GPU'dan bellek/enerjiye genisledi.** Memory shortage, data center emissions, edge inference ve distributed LLM basliklari compute stratejisinin tek boyutlu GPU alimi olmadigini gosteriyor. Kaynak: https://www.theregister.com/ai-and-ml/2026/07/12/memory-makers-are-slaves-to-the-boom-bust-rollercoaster-and-the-ai-boom-is-the-wildest-ride-of-all/5269549

## Kaynak Ozeti

- Raw: 5.027 item.
- Merge: 4.992 item, 35 URL tekrari atildi.
- Dedupe: 1.640 yeni, 3.347 fingerprint tekrari, 5 semantik tekrar, 0 materyal guncelleme.
- Top source family: social 4.133 raw / 1.213 unique.
- Top unique kaynaklar: google_news/ai 69, google_news/security 61, google_news/companies 52, google_news/funding 50, twitter_fallback/@drfeifei 39.

## Dedupe & Kalite Notu

> Bu rapordaki maddeler v3 dedupe pipeline'i ile onceki state ve bugunku merged item havuzuna karsi elendi. Toplam 4.992 merged item islendi, 1.640 benzersiz item rapor havuzuna girdi.

Kalite notu: Collector, merge, dedupe ve trend adimlari basariyla calisti. Rapor uretiminde otomatik Codex LLM cagrisi 401 Unauthorized verdigi icin bu dosya manuel fallback senteziyle olusturuldu; her madde yine unique.json ve collector kaynak URL'lerine dayandirildi.