Japonya, siber güvenliklerini artırmak amacıyla OpenAI ve Anthropic ile anlaşmalar imzaladı. Ülkenin finans kuruluşları, OpenAI'ın GPT-5.5 Cyber sürümünü ve Anthropic'in Mythos modelini kullanacak.
Japonya, finans sektörünü siber saldırılardan korumak için OpenAI'ın GPT-5.5 modelinin özel bir sürümünü kullanmaya başlıyor. Ülkenin büyük bankaları bu teknolojiye erişecek.
A new benchmark, Claw-Anything, tests AI personal assistants on realistic, long-horizon tasks across multiple devices and services, revealing poor performance (e.g., GPT-5.5 scored 34.5%) due to complexity and noise in simulated user activity.
Advanced AI models like Anthropic's Mythos and GPT-5.5 demonstrate sophisticated cyber capabilities, finding thousands of high-severity vulnerabilities and outpacing human experts. This rapid advancement raises concerns about potential misuse by adversaries, prompting calls for government access to secure critical networks.
米オープンAIは、サイバー防御に特化した最新モデル「GPT5.5サイバー」を日本の政府や企業向けに提供する方針を発表。元米陸軍大将で同社取締役のポール・ナカソネ氏が、自由で開かれたインド太平洋（FOIP）における日本の重要性を強調し、早期提供に向けた政府との協議を進めると述べた。
OpenAI, Daybreak adında yeni bir siber güvenlik girişimini duyurdu. Bu girişim, siber savunma süreçlerini yapay zeka modelleriyle daha hızlı ve verimli hale getirmeyi hedefliyor.
OpenAI, sesli yapay zeka teknolojilerinde çıtayı bambaşka bir noktaya taşıyan yeni modellerini resmen duyurdu. Geliştiriciler için yayınlanan GPT-Realtime-2, GPT-Realtime-Translate ve GPT-Realtime-Whisper isimli bu üç yeni model, sesli etkileşimleri sadece bir soru-cevap döngüsü olmaktan çıkarıp gerçek zamanlı düşünebilen, çeviri yapabilen ve karmaşık görevleri yerine getirebilen dijital asistanlara dönüştürüyor. Özellikle GPT-Realtime-2, sahip olduğu GPT-5 sınıfı akıl yürütme […]
OpenAI said GPT-5.5-Cyber, a variation of its latest AI model, is rolling out in a limited preview capacity to vetted cybersecurity teams.
Das neue Standardmodell GPT-5.5 Instant soll laut OpenAI weniger halluzinieren, prägnantere Antworten liefern und stärker persönlichen Kontext einbeziehen.
GPT-5.5 Instant is now the default ChatGPT model. It makes fewer things up, answers more concisely, and finally remembers what you told it last week.
OpenAI is upgrading ChatGPT's default model to GPT-5.5 Instant, touting reduced hallucinations and more direct responses. A new 'memory sources' feature will offer users more control over personalization data.
Teknoloji dünyasının öncü yapay zeka şirketi OpenAI, pazartesi günü yaptığı duyuruyla ChatGPT’nin yeni varsayılan temel modeli olan GPT-5.5 Instant‘ı resmen kullanıma sunduğunu açıkladı. Bu yeni sürüm, selefi olan GPT-5.3 Instant’ın yerini alarak platformun standart deneyimi haline geliyor. Hukuk ve Tıpta Daha Az Halüsinasyon, Daha Yüksek Başarı Şirketin açıklamalarına göre GPT-5.5 Instant, önceki sürümün sahip olduğu […]
OpenAI's newest default model for ChatGPT might not make stuff up as much. Hallucinations have been an ongoing problem for AI models, but OpenAI says its new GPT-5.5 Instant model has "significant improvements in factuality across the board." The company claims that, based on "internal evaluations," GPT-5.5 Instant produced "52.5% fewer hallucinated claims" than its […]
The new GPT-5.5 Instant model will replace GPT-3.5 Instant as the default model for ChatGPT
NIST's CAISI evaluated DeepSeek V4 Pro using private benchmarks and a cost-comparison filter that excluded every US model except GPT-5.4 mini. Critics call the methodology convenient.
OpenAI CEO Sam Altman has publicly invited Elon Musk to attend the company’s private GPT-5.5 event, despite an ongoing high-profile lawsuit between the two parties regarding OpenAI's shift from a non-profit mission to a for-profit model.
AISI testing shows GPT-5.5 completed a 32-step corporate network attack simulation autonomously and solved a reverse-engineering challenge in 10 minutes that took a human 12 hours. The model achieved 71.4% pass rate on Expert-tier cybersecurity tasks, edging out Anthropic's Claude Mythos. Researchers also identified a universal jailbreak affecting all malicious queries tested.
The UK's AI Security Institute released research showing OpenAI's GPT-5.5 achieved 71.4% on Expert-level cybersecurity tasks versus Anthropic's Mythos Preview at 68.6%—within margin of error. Both models succeeded on the previously impossible "The Last Ones" test (GPT-5.5: 3/10, Mythos: 2/10), though both failed the "Cooling Tower" power plant simulation. Sam Altman criticized Anthropic's restricted release as "fear-based marketing," suggesting the capabilities reflect general AI improvements rather than a model-specific breakthrough.
OpenAI discovered ChatGPT developed an unusual obsession with goblins, gremlins, raccoons, trolls, and pigeons after a 'Nerdy' personality setting inadvertently rewarded such language in training. The Nerdy mode, representing just 2.5% of responses, accounted for 66.7% of all goblin mentions. Usage of 'goblin' rose 175% after GPT-5.1, with the reward signal scoring goblin-containing outputs higher in 76.2% of datasets. OpenAI retired the Nerdy personality in March 2026 and removed the goblin-friendly reward signal.
OpenAI will begin rolling out its cybersecurity tool GPT-5.5 Cyber to critical cyber defenders in the coming days, CEO Sam Altman announced on X. The tool can perform penetration testing, vulnerability identification and exploitation, and malware reverse engineering. Altman had previously criticized Anthropic for restricting access to its Mythos tool, calling it fear-based marketing, but now OpenAI is implementing the same restricted-access model.
OpenAI added a system prompt to GPT-5.5's Codex app to avoid mentioning goblins and other creatures, following a surge in their use by ChatGPT. The company traced the 'goblin obsession' to a 'nerdy' personality feature that disproportionately used the word, with the behavior transferring to other model parts via reinforcement learning.
OpenAI discovered a 175% increase in ChatGPT mentioning goblins and gremlins after GPT-5.1's November launch, traced to a retired 'nerdy personality' training that inadvertently incentivized creature references. The company updated Codex to avoid goblins, gremlins, raccoons, trolls, ogres, and pigeons unless relevant, with the personality responsible for 66.7% of goblin mentions.
OpenAI announced it will launch GPT-5.5-Cyber, a specialized cybersecurity AI model, to a select group of trusted 'cyber defenders' in the next few days. CEO Sam Altman stated the model will not be available to the general public, with access limited to vetted institutions to allow them to strengthen their cyberdefenses. The company plans to work with the ecosystem and government to determine trusted access parameters. The staggered release follows a growing industry trend of restricting access to powerful AI models due to misuse concerns, similar to Anthropic's Claude Mythos release which faced White House opposition.
OpenAI's recently released Codex CLI contains a puzzling system prompt warning for GPT-5.5 to 'never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query.' The prohibition appears twice in the base instructions. OpenAI employee Nick Pash insisted the warning 'isn't a marketing gimmick,' while CEO Sam Altman joked that 'codex is having a goblin moment.' The issue mirrors a similar problem xAI's Grok had last year with bringing up 'white genocide' in unrelated conversations.
