Given the various ways in which AI language models can be misused and the absence of a known fix, what steps are being taken by the tech companies to mitigate these risks? Have there been any recent developments in this area that could help prevent malicious attacks on these models, and what role can researchers play in identifying potential vulnerabilities before they are exploited?