Konstantinos Daskalakis warns of real risks from unpredictable AI behavior

AI agents unpredictable behavior artificial intelligence safety neural networks technological regulation

Konstantinos Daskalakis, an MIT professor and former chair of the AI Advisory Committee, warns that while the total loss of control over AI remains a hypothetical risk, the danger of unpredictable or undesirable behavior is already a present reality. This unpredictability is inherent to the training and operation of large neural networks.

The discussion has shifted from simple chatbots to advanced AI agents—systems capable of planning, executing actions, using digital tools, and writing code with minimal human intervention. These agents pose significant risks, particularly regarding cyberattacks and autonomous actions.

In the broader industry, leaders are divided. Anthropic CEO Dario Amodei has proposed slowing technological progress to ensure public safety, a move supported by Sam Altman of OpenAI, Demis Hassabis of Google DeepMind, and Elon Musk of xAI. Meanwhile, Microsoft co-founder Bill Gates has cautioned that no government is currently prepared for the societal shifts AI will trigger, including job displacement and cyber threats.

Entities

Anthropic · Bill Gates · European Union · Konstantinos Daskalakis · MIT · OpenAI · Silicon Valley

Claims

What the coverage asserts, and how many sources carry each claim.

[● 8 SOURCES] AI agents are systems capable of planning, executing actions, and using digital tools with limited human intervention. athina984.gr · www.iefimerida.gr · www.in.gr · www.lamiareport.gr · www.anatolh.com · +3 more
[● 8 SOURCES] Unpredictable behavior is inherent to the way large neural networks are trained and operate. athina984.gr · www.in.gr · www.lamiareport.gr · www.anatolh.com · latofm.gr · +3 more
[DISPUTED] The risk of losing total control over AI systems remains hypothetical. athina984.gr · www.iefimerida.gr · www.in.gr · www.lamiareport.gr · www.anatolh.com · +3 more
[○ 1 SOURCE] Cyberattacks by autonomous AI agents represent an immediate risk. www.iefimerida.gr
[DISPUTED] The risk of unpredictable or undesirable behavior in advanced AI models is a present reality. athina984.gr · www.iefimerida.gr · www.in.gr · www.lamiareport.gr · www.anatolh.com · +3 more
[● 8 SOURCES] Reliability certification for generative AI and large language models may never be possible. athina984.gr · www.in.gr · www.lamiareport.gr · www.anatolh.com · latofm.gr · +3 more
[● 2 SOURCES] Microsoft co-founder Bill Gates stated that no government is prepared for the societal changes brought by AI. www.gargalianoionline.gr · www.in.gr
[● 2 SOURCES] Anthropic CEO Dario Amodei proposed slowing down technological progress to ensure public safety. www.gargalianoionline.gr · www.in.gr
[● 2 SOURCES] Anthropic researchers have warned that AI could wipe out humanity by 2030. www.gargalianoionline.gr · www.in.gr