AIが期間内の動向を整理
AI news roundup: faster document work, longer-running agents, and a rethink of safety design
AI news on July 17 was marked by announcements from OpenAI, Google, and Anthropic. The updates center on faster document work, agent infrastructure that can keep running for longer jobs, and clearer thinking about safety design and risky misuse.
SOURCE CHECK
Primary Sources 111
Primary Sources
SOURCE CHECK 1
openai.com
SOURCE CHECK 2
openai.com
SOURCE CHECK 3
openai.com
SOURCE CHECK 4
openai.com
SOURCE CHECK 5
openai.com
SOURCE CHECK 6
openai.com
SOURCE CHECK 7
openai.com
SOURCE CHECK 8
openai.com
SOURCE CHECK 9
deploymentsafety.openai.com
SOURCE CHECK 10
openai.com
SOURCE CHECK 11
openai.com
SOURCE CHECK 12
x.com
SOURCE CHECK 13
openai.com
SOURCE CHECK 14
openai.com
SOURCE CHECK 15
x.com
SOURCE CHECK 16
x.com
SOURCE CHECK 17
openai.com
SOURCE CHECK 18
openai.com
SOURCE CHECK 19
openai.com
SOURCE CHECK 20
openai.com
SOURCE CHECK 21
openai.com
SOURCE CHECK 22
x.com
SOURCE CHECK 23
openai.com
SOURCE CHECK 24
x.com
SOURCE CHECK 25
x.com
SOURCE CHECK 26
x.com
SOURCE CHECK 27
x.com
SOURCE CHECK 28
x.com
SOURCE CHECK 29
openai.com
SOURCE CHECK 30
openai.com
SOURCE CHECK 31
openai.com
SOURCE CHECK 32
openai.com
SOURCE CHECK 33
openai.com
SOURCE CHECK 34
openai.com
SOURCE CHECK 35
openai.com
SOURCE CHECK 36
openai.com
SOURCE CHECK 37
openai.com
SOURCE CHECK 38
openai.com
SOURCE CHECK 39
developers.openai.com
SOURCE CHECK 40
x.com
SOURCE CHECK 41
openai.com
SOURCE CHECK 42
openai.com
SOURCE CHECK 43
openai.com
SOURCE CHECK 44
ai.google.dev
SOURCE CHECK 45
ai.google.dev
SOURCE CHECK 46
blog.google
SOURCE CHECK 47
ai.google.dev
SOURCE CHECK 48
blog.google
SOURCE CHECK 49
blog.google
SOURCE CHECK 50
blog.google
SOURCE CHECK 51
blog.google
SOURCE CHECK 52
blog.google
SOURCE CHECK 53
blog.google
SOURCE CHECK 54
blog.google
SOURCE CHECK 55
ai.google.dev
SOURCE CHECK 56
blog.google
SOURCE CHECK 57
blog.google
SOURCE CHECK 58
blog.google
SOURCE CHECK 59
blog.google
SOURCE CHECK 60
blog.google
SOURCE CHECK 61
developers.googleblog.com
SOURCE CHECK 62
blog.google
SOURCE CHECK 63
blog.google
SOURCE CHECK 64
blog.google
SOURCE CHECK 65
blog.google
SOURCE CHECK 66
blog.google
SOURCE CHECK 67
blog.google
SOURCE CHECK 68
blog.google
SOURCE CHECK 69
blog.google
SOURCE CHECK 70
blog.google
SOURCE CHECK 71
developers.googleblog.com
SOURCE CHECK 72
developers.googleblog.com
SOURCE CHECK 73
blog.google
SOURCE CHECK 74
blog.google
SOURCE CHECK 75
blog.google
SOURCE CHECK 76
blog.google
SOURCE CHECK 77
blog.google
SOURCE CHECK 78
blog.google
SOURCE CHECK 79
blog.google
SOURCE CHECK 80
x.com
SOURCE CHECK 81
ai.google.dev
SOURCE CHECK 82
ai.google.dev
SOURCE CHECK 83
blog.google
SOURCE CHECK 84
deepmind.google
SOURCE CHECK 85
ai.google.dev
SOURCE CHECK 86
ai.google.dev
SOURCE CHECK 87
blog.google
SOURCE CHECK 88
x.com
SOURCE CHECK 89
blog.google
SOURCE CHECK 90
blog.google
SOURCE CHECK 91
blog.google
SOURCE CHECK 92
blog.google
SOURCE CHECK 93
ai.google.dev
SOURCE CHECK 94
ai.google.dev
SOURCE CHECK 95
anthropic.com
SOURCE CHECK 96
anthropic.com
SOURCE CHECK 97
anthropic.com
SOURCE CHECK 98
anthropic.com
SOURCE CHECK 99
anthropic.com
SOURCE CHECK 100
anthropic.com
SOURCE CHECK 101
anthropic.com
SOURCE CHECK 102
anthropic.com
SOURCE CHECK 103
anthropic.com
SOURCE CHECK 104
anthropic.com
SOURCE CHECK 105
anthropic.com
SOURCE CHECK 106
anthropic.com
SOURCE CHECK 107
anthropic.com
SOURCE CHECK 108
anthropic.com
SOURCE CHECK 109
anthropic.com
SOURCE CHECK 110
x.com
SOURCE CHECK 111
anthropic.com
Key Points
- 1OpenAI generally released GPT-5.6 to make it easier to move from conversation to documents and code work
- 2Google upgraded Managed Agents in the Gemini API to make long-running processing and external connections easier
- 3Anthropic published details on Fable 5 cyber safeguards and a framework for judging jailbreak severity
- 4Anthropic also released Claude for Teachers, emphasizing lesson-prep automation and privacy protection
OpenAI: making the path from conversation to documents more seamless
OpenAI generally released GPT-5.6, making it easier to use the model across conversation, research, document creation, and coding. The API offers three sizes: Sol, Terra, and Luna, which makes it easier to choose the right balance of performance and cost for different tasks. For day-to-day work, it supports a workflow where you can start with a discussion and continue directly into a document.
Google: easier long-running AI tasks and external connections
Google strengthened Managed Agents in the Gemini API with background execution, remote MCP server integration, and custom function calling. That reduces the chance that automated tasks stop halfway through and makes it easier to coordinate work across internal data and external services. For developers, it is a step closer to practical agent design.
Anthropic: clarifying how safety measures are presented
Anthropic published details on Fable 5 cyber safeguards and a framework for judging the severity of jailbreaks. It explains the role of classifiers that help prevent dangerous misuse, and it provides a useful basis for discussion among researchers and other companies. For readers comparing models, it is important not just for performance but for safety design.
Anthropic’s education-focused Claude: automation with privacy considerations
Anthropic released Claude for Teachers for K-12 education. It generates lesson plans and materials based on state standards and existing curricula. Student data is FERPA-compliant and conversations are not used for training. For education settings, the key decision point is not only convenience but also data protection.